4 This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
6 Copyright (C) 2001-2008, LINBIT Information Technologies GmbH.
7 Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>.
8 Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.
10 drbd is free software; you can redistribute it and/or modify
11 it under the terms of the GNU General Public License as published by
12 the Free Software Foundation; either version 2, or (at your option)
15 drbd is distributed in the hope that it will be useful,
16 but WITHOUT ANY WARRANTY; without even the implied warranty of
17 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 GNU General Public License for more details.
20 You should have received a copy of the GNU General Public License
21 along with drbd; see the file COPYING. If not, write to
22 the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
26 #include <linux/module.h>
28 #include <asm/uaccess.h>
31 #include <linux/drbd.h>
33 #include <linux/file.h>
36 #include <linux/memcontrol.h>
37 #include <linux/mm_inline.h>
38 #include <linux/slab.h>
39 #include <linux/pkt_sched.h>
40 #define __KERNEL_SYSCALLS__
41 #include <linux/unistd.h>
42 #include <linux/vmalloc.h>
43 #include <linux/random.h>
44 #include <linux/string.h>
45 #include <linux/scatterlist.h>
47 #include "drbd_protocol.h"
65 static int drbd_do_features(struct drbd_connection *connection);
66 static int drbd_do_auth(struct drbd_connection *connection);
67 static int drbd_disconnected(struct drbd_device *device);
69 static enum finish_epoch drbd_may_finish_epoch(struct drbd_connection *, struct drbd_epoch *, enum epoch_event);
70 static int e_end_block(struct drbd_work *, int);
73 #define GFP_TRY (__GFP_HIGHMEM | __GFP_NOWARN)
76 * some helper functions to deal with single linked page lists,
77 * page->private being our "next" pointer.
80 /* If at least n pages are linked at head, get n pages off.
81 * Otherwise, don't modify head, and return NULL.
82 * Locking is the responsibility of the caller.
84 static struct page *page_chain_del(struct page **head, int n)
98 tmp = page_chain_next(page);
100 break; /* found sufficient pages */
102 /* insufficient pages, don't use any of them. */
107 /* add end of list marker for the returned list */
108 set_page_private(page, 0);
109 /* actual return value, and adjustment of head */
115 /* may be used outside of locks to find the tail of a (usually short)
116 * "private" page chain, before adding it back to a global chain head
117 * with page_chain_add() under a spinlock. */
118 static struct page *page_chain_tail(struct page *page, int *len)
122 while ((tmp = page_chain_next(page)))
129 static int page_chain_free(struct page *page)
133 page_chain_for_each_safe(page, tmp) {
140 static void page_chain_add(struct page **head,
141 struct page *chain_first, struct page *chain_last)
145 tmp = page_chain_tail(chain_first, NULL);
146 BUG_ON(tmp != chain_last);
149 /* add chain to head */
150 set_page_private(chain_last, (unsigned long)*head);
154 static struct page *__drbd_alloc_pages(struct drbd_device *device,
157 struct page *page = NULL;
158 struct page *tmp = NULL;
161 /* Yes, testing drbd_pp_vacant outside the lock is racy.
162 * So what. It saves a spin_lock. */
163 if (drbd_pp_vacant >= number) {
164 spin_lock(&drbd_pp_lock);
165 page = page_chain_del(&drbd_pp_pool, number);
167 drbd_pp_vacant -= number;
168 spin_unlock(&drbd_pp_lock);
173 /* GFP_TRY, because we must not cause arbitrary write-out: in a DRBD
174 * "criss-cross" setup, that might cause write-out on some other DRBD,
175 * which in turn might block on the other node at this very place. */
176 for (i = 0; i < number; i++) {
177 tmp = alloc_page(GFP_TRY);
180 set_page_private(tmp, (unsigned long)page);
187 /* Not enough pages immediately available this time.
188 * No need to jump around here, drbd_alloc_pages will retry this
189 * function "soon". */
191 tmp = page_chain_tail(page, NULL);
192 spin_lock(&drbd_pp_lock);
193 page_chain_add(&drbd_pp_pool, page, tmp);
195 spin_unlock(&drbd_pp_lock);
200 static void reclaim_finished_net_peer_reqs(struct drbd_device *device,
201 struct list_head *to_be_freed)
203 struct drbd_peer_request *peer_req;
204 struct list_head *le, *tle;
206 /* The EEs are always appended to the end of the list. Since
207 they are sent in order over the wire, they have to finish
208 in order. As soon as we see the first not finished we can
209 stop to examine the list... */
211 list_for_each_safe(le, tle, &device->net_ee) {
212 peer_req = list_entry(le, struct drbd_peer_request, w.list);
213 if (drbd_peer_req_has_active_page(peer_req))
215 list_move(le, to_be_freed);
219 static void drbd_kick_lo_and_reclaim_net(struct drbd_device *device)
221 LIST_HEAD(reclaimed);
222 struct drbd_peer_request *peer_req, *t;
224 spin_lock_irq(&first_peer_device(device)->connection->req_lock);
225 reclaim_finished_net_peer_reqs(device, &reclaimed);
226 spin_unlock_irq(&first_peer_device(device)->connection->req_lock);
228 list_for_each_entry_safe(peer_req, t, &reclaimed, w.list)
229 drbd_free_net_peer_req(device, peer_req);
233 * drbd_alloc_pages() - Returns @number pages, retries forever (or until signalled)
234 * @device: DRBD device.
235 * @number: number of pages requested
236 * @retry: whether to retry, if not enough pages are available right now
238 * Tries to allocate number pages, first from our own page pool, then from
239 * the kernel, unless this allocation would exceed the max_buffers setting.
240 * Possibly retry until DRBD frees sufficient pages somewhere else.
242 * Returns a page chain linked via page->private.
244 struct page *drbd_alloc_pages(struct drbd_device *device, unsigned int number,
247 struct page *page = NULL;
252 /* Yes, we may run up to @number over max_buffers. If we
253 * follow it strictly, the admin will get it wrong anyways. */
255 nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
256 mxb = nc ? nc->max_buffers : 1000000;
259 if (atomic_read(&device->pp_in_use) < mxb)
260 page = __drbd_alloc_pages(device, number);
262 while (page == NULL) {
263 prepare_to_wait(&drbd_pp_wait, &wait, TASK_INTERRUPTIBLE);
265 drbd_kick_lo_and_reclaim_net(device);
267 if (atomic_read(&device->pp_in_use) < mxb) {
268 page = __drbd_alloc_pages(device, number);
276 if (signal_pending(current)) {
277 dev_warn(DEV, "drbd_alloc_pages interrupted!\n");
283 finish_wait(&drbd_pp_wait, &wait);
286 atomic_add(number, &device->pp_in_use);
290 /* Must not be used from irq, as that may deadlock: see drbd_alloc_pages.
291 * Is also used from inside an other spin_lock_irq(&first_peer_device(device)->connection->req_lock);
292 * Either links the page chain back to the global pool,
293 * or returns all pages to the system. */
294 static void drbd_free_pages(struct drbd_device *device, struct page *page, int is_net)
296 atomic_t *a = is_net ? &device->pp_in_use_by_net : &device->pp_in_use;
302 if (drbd_pp_vacant > (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * minor_count)
303 i = page_chain_free(page);
306 tmp = page_chain_tail(page, &i);
307 spin_lock(&drbd_pp_lock);
308 page_chain_add(&drbd_pp_pool, page, tmp);
310 spin_unlock(&drbd_pp_lock);
312 i = atomic_sub_return(i, a);
314 dev_warn(DEV, "ASSERTION FAILED: %s: %d < 0\n",
315 is_net ? "pp_in_use_by_net" : "pp_in_use", i);
316 wake_up(&drbd_pp_wait);
320 You need to hold the req_lock:
321 _drbd_wait_ee_list_empty()
323 You must not have the req_lock:
325 drbd_alloc_peer_req()
326 drbd_free_peer_reqs()
328 drbd_finish_peer_reqs()
330 drbd_wait_ee_list_empty()
333 struct drbd_peer_request *
334 drbd_alloc_peer_req(struct drbd_device *device, u64 id, sector_t sector,
335 unsigned int data_size, gfp_t gfp_mask) __must_hold(local)
337 struct drbd_peer_request *peer_req;
338 struct page *page = NULL;
339 unsigned nr_pages = (data_size + PAGE_SIZE -1) >> PAGE_SHIFT;
341 if (drbd_insert_fault(device, DRBD_FAULT_AL_EE))
344 peer_req = mempool_alloc(drbd_ee_mempool, gfp_mask & ~__GFP_HIGHMEM);
346 if (!(gfp_mask & __GFP_NOWARN))
347 dev_err(DEV, "%s: allocation failed\n", __func__);
352 page = drbd_alloc_pages(device, nr_pages, (gfp_mask & __GFP_WAIT));
357 drbd_clear_interval(&peer_req->i);
358 peer_req->i.size = data_size;
359 peer_req->i.sector = sector;
360 peer_req->i.local = false;
361 peer_req->i.waiting = false;
363 peer_req->epoch = NULL;
364 peer_req->w.device = device;
365 peer_req->pages = page;
366 atomic_set(&peer_req->pending_bios, 0);
369 * The block_id is opaque to the receiver. It is not endianness
370 * converted, and sent back to the sender unchanged.
372 peer_req->block_id = id;
377 mempool_free(peer_req, drbd_ee_mempool);
381 void __drbd_free_peer_req(struct drbd_device *device, struct drbd_peer_request *peer_req,
384 if (peer_req->flags & EE_HAS_DIGEST)
385 kfree(peer_req->digest);
386 drbd_free_pages(device, peer_req->pages, is_net);
387 D_ASSERT(atomic_read(&peer_req->pending_bios) == 0);
388 D_ASSERT(drbd_interval_empty(&peer_req->i));
389 mempool_free(peer_req, drbd_ee_mempool);
392 int drbd_free_peer_reqs(struct drbd_device *device, struct list_head *list)
394 LIST_HEAD(work_list);
395 struct drbd_peer_request *peer_req, *t;
397 int is_net = list == &device->net_ee;
399 spin_lock_irq(&first_peer_device(device)->connection->req_lock);
400 list_splice_init(list, &work_list);
401 spin_unlock_irq(&first_peer_device(device)->connection->req_lock);
403 list_for_each_entry_safe(peer_req, t, &work_list, w.list) {
404 __drbd_free_peer_req(device, peer_req, is_net);
411 * See also comments in _req_mod(,BARRIER_ACKED) and receive_Barrier.
413 static int drbd_finish_peer_reqs(struct drbd_device *device)
415 LIST_HEAD(work_list);
416 LIST_HEAD(reclaimed);
417 struct drbd_peer_request *peer_req, *t;
420 spin_lock_irq(&first_peer_device(device)->connection->req_lock);
421 reclaim_finished_net_peer_reqs(device, &reclaimed);
422 list_splice_init(&device->done_ee, &work_list);
423 spin_unlock_irq(&first_peer_device(device)->connection->req_lock);
425 list_for_each_entry_safe(peer_req, t, &reclaimed, w.list)
426 drbd_free_net_peer_req(device, peer_req);
428 /* possible callbacks here:
429 * e_end_block, and e_end_resync_block, e_send_superseded.
430 * all ignore the last argument.
432 list_for_each_entry_safe(peer_req, t, &work_list, w.list) {
435 /* list_del not necessary, next/prev members not touched */
436 err2 = peer_req->w.cb(&peer_req->w, !!err);
439 drbd_free_peer_req(device, peer_req);
441 wake_up(&device->ee_wait);
446 static void _drbd_wait_ee_list_empty(struct drbd_device *device,
447 struct list_head *head)
451 /* avoids spin_lock/unlock
452 * and calling prepare_to_wait in the fast path */
453 while (!list_empty(head)) {
454 prepare_to_wait(&device->ee_wait, &wait, TASK_UNINTERRUPTIBLE);
455 spin_unlock_irq(&first_peer_device(device)->connection->req_lock);
457 finish_wait(&device->ee_wait, &wait);
458 spin_lock_irq(&first_peer_device(device)->connection->req_lock);
462 static void drbd_wait_ee_list_empty(struct drbd_device *device,
463 struct list_head *head)
465 spin_lock_irq(&first_peer_device(device)->connection->req_lock);
466 _drbd_wait_ee_list_empty(device, head);
467 spin_unlock_irq(&first_peer_device(device)->connection->req_lock);
470 static int drbd_recv_short(struct socket *sock, void *buf, size_t size, int flags)
477 struct msghdr msg = {
479 .msg_iov = (struct iovec *)&iov,
480 .msg_flags = (flags ? flags : MSG_WAITALL | MSG_NOSIGNAL)
486 rv = sock_recvmsg(sock, &msg, size, msg.msg_flags);
492 static int drbd_recv(struct drbd_connection *connection, void *buf, size_t size)
496 rv = drbd_recv_short(connection->data.socket, buf, size, 0);
499 if (rv == -ECONNRESET)
500 conn_info(connection, "sock was reset by peer\n");
501 else if (rv != -ERESTARTSYS)
502 conn_err(connection, "sock_recvmsg returned %d\n", rv);
503 } else if (rv == 0) {
504 if (test_bit(DISCONNECT_SENT, &connection->flags)) {
507 t = rcu_dereference(connection->net_conf)->ping_timeo * HZ/10;
510 t = wait_event_timeout(connection->ping_wait, connection->cstate < C_WF_REPORT_PARAMS, t);
515 conn_info(connection, "sock was shut down by peer\n");
519 conn_request_state(connection, NS(conn, C_BROKEN_PIPE), CS_HARD);
525 static int drbd_recv_all(struct drbd_connection *connection, void *buf, size_t size)
529 err = drbd_recv(connection, buf, size);
538 static int drbd_recv_all_warn(struct drbd_connection *connection, void *buf, size_t size)
542 err = drbd_recv_all(connection, buf, size);
543 if (err && !signal_pending(current))
544 conn_warn(connection, "short read (expected size %d)\n", (int)size);
549 * On individual connections, the socket buffer size must be set prior to the
550 * listen(2) or connect(2) calls in order to have it take effect.
551 * This is our wrapper to do so.
553 static void drbd_setbufsize(struct socket *sock, unsigned int snd,
556 /* open coded SO_SNDBUF, SO_RCVBUF */
558 sock->sk->sk_sndbuf = snd;
559 sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
562 sock->sk->sk_rcvbuf = rcv;
563 sock->sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
567 static struct socket *drbd_try_connect(struct drbd_connection *connection)
571 struct sockaddr_in6 src_in6;
572 struct sockaddr_in6 peer_in6;
574 int err, peer_addr_len, my_addr_len;
575 int sndbuf_size, rcvbuf_size, connect_int;
576 int disconnect_on_error = 1;
579 nc = rcu_dereference(connection->net_conf);
584 sndbuf_size = nc->sndbuf_size;
585 rcvbuf_size = nc->rcvbuf_size;
586 connect_int = nc->connect_int;
589 my_addr_len = min_t(int, connection->my_addr_len, sizeof(src_in6));
590 memcpy(&src_in6, &connection->my_addr, my_addr_len);
592 if (((struct sockaddr *)&connection->my_addr)->sa_family == AF_INET6)
593 src_in6.sin6_port = 0;
595 ((struct sockaddr_in *)&src_in6)->sin_port = 0; /* AF_INET & AF_SCI */
597 peer_addr_len = min_t(int, connection->peer_addr_len, sizeof(src_in6));
598 memcpy(&peer_in6, &connection->peer_addr, peer_addr_len);
600 what = "sock_create_kern";
601 err = sock_create_kern(((struct sockaddr *)&src_in6)->sa_family,
602 SOCK_STREAM, IPPROTO_TCP, &sock);
608 sock->sk->sk_rcvtimeo =
609 sock->sk->sk_sndtimeo = connect_int * HZ;
610 drbd_setbufsize(sock, sndbuf_size, rcvbuf_size);
612 /* explicitly bind to the configured IP as source IP
613 * for the outgoing connections.
614 * This is needed for multihomed hosts and to be
615 * able to use lo: interfaces for drbd.
616 * Make sure to use 0 as port number, so linux selects
617 * a free one dynamically.
619 what = "bind before connect";
620 err = sock->ops->bind(sock, (struct sockaddr *) &src_in6, my_addr_len);
624 /* connect may fail, peer not yet available.
625 * stay C_WF_CONNECTION, don't go Disconnecting! */
626 disconnect_on_error = 0;
628 err = sock->ops->connect(sock, (struct sockaddr *) &peer_in6, peer_addr_len, 0);
637 /* timeout, busy, signal pending */
638 case ETIMEDOUT: case EAGAIN: case EINPROGRESS:
639 case EINTR: case ERESTARTSYS:
640 /* peer not (yet) available, network problem */
641 case ECONNREFUSED: case ENETUNREACH:
642 case EHOSTDOWN: case EHOSTUNREACH:
643 disconnect_on_error = 0;
646 conn_err(connection, "%s failed, err = %d\n", what, err);
648 if (disconnect_on_error)
649 conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD);
655 struct accept_wait_data {
656 struct drbd_connection *connection;
657 struct socket *s_listen;
658 struct completion door_bell;
659 void (*original_sk_state_change)(struct sock *sk);
663 static void drbd_incoming_connection(struct sock *sk)
665 struct accept_wait_data *ad = sk->sk_user_data;
666 void (*state_change)(struct sock *sk);
668 state_change = ad->original_sk_state_change;
669 if (sk->sk_state == TCP_ESTABLISHED)
670 complete(&ad->door_bell);
674 static int prepare_listen_socket(struct drbd_connection *connection, struct accept_wait_data *ad)
676 int err, sndbuf_size, rcvbuf_size, my_addr_len;
677 struct sockaddr_in6 my_addr;
678 struct socket *s_listen;
683 nc = rcu_dereference(connection->net_conf);
688 sndbuf_size = nc->sndbuf_size;
689 rcvbuf_size = nc->rcvbuf_size;
692 my_addr_len = min_t(int, connection->my_addr_len, sizeof(struct sockaddr_in6));
693 memcpy(&my_addr, &connection->my_addr, my_addr_len);
695 what = "sock_create_kern";
696 err = sock_create_kern(((struct sockaddr *)&my_addr)->sa_family,
697 SOCK_STREAM, IPPROTO_TCP, &s_listen);
703 s_listen->sk->sk_reuse = SK_CAN_REUSE; /* SO_REUSEADDR */
704 drbd_setbufsize(s_listen, sndbuf_size, rcvbuf_size);
706 what = "bind before listen";
707 err = s_listen->ops->bind(s_listen, (struct sockaddr *)&my_addr, my_addr_len);
711 ad->s_listen = s_listen;
712 write_lock_bh(&s_listen->sk->sk_callback_lock);
713 ad->original_sk_state_change = s_listen->sk->sk_state_change;
714 s_listen->sk->sk_state_change = drbd_incoming_connection;
715 s_listen->sk->sk_user_data = ad;
716 write_unlock_bh(&s_listen->sk->sk_callback_lock);
719 err = s_listen->ops->listen(s_listen, 5);
726 sock_release(s_listen);
728 if (err != -EAGAIN && err != -EINTR && err != -ERESTARTSYS) {
729 conn_err(connection, "%s failed, err = %d\n", what, err);
730 conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD);
737 static void unregister_state_change(struct sock *sk, struct accept_wait_data *ad)
739 write_lock_bh(&sk->sk_callback_lock);
740 sk->sk_state_change = ad->original_sk_state_change;
741 sk->sk_user_data = NULL;
742 write_unlock_bh(&sk->sk_callback_lock);
745 static struct socket *drbd_wait_for_connect(struct drbd_connection *connection, struct accept_wait_data *ad)
747 int timeo, connect_int, err = 0;
748 struct socket *s_estab = NULL;
752 nc = rcu_dereference(connection->net_conf);
757 connect_int = nc->connect_int;
760 timeo = connect_int * HZ;
761 /* 28.5% random jitter */
762 timeo += (prandom_u32() & 1) ? timeo / 7 : -timeo / 7;
764 err = wait_for_completion_interruptible_timeout(&ad->door_bell, timeo);
768 err = kernel_accept(ad->s_listen, &s_estab, 0);
770 if (err != -EAGAIN && err != -EINTR && err != -ERESTARTSYS) {
771 conn_err(connection, "accept failed, err = %d\n", err);
772 conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD);
777 unregister_state_change(s_estab->sk, ad);
782 static int decode_header(struct drbd_connection *, void *, struct packet_info *);
784 static int send_first_packet(struct drbd_connection *connection, struct drbd_socket *sock,
785 enum drbd_packet cmd)
787 if (!conn_prepare_command(connection, sock))
789 return conn_send_command(connection, sock, cmd, 0, NULL, 0);
792 static int receive_first_packet(struct drbd_connection *connection, struct socket *sock)
794 unsigned int header_size = drbd_header_size(connection);
795 struct packet_info pi;
798 err = drbd_recv_short(sock, connection->data.rbuf, header_size, 0);
799 if (err != header_size) {
804 err = decode_header(connection, connection->data.rbuf, &pi);
811 * drbd_socket_okay() - Free the socket if its connection is not okay
812 * @sock: pointer to the pointer to the socket.
814 static int drbd_socket_okay(struct socket **sock)
822 rr = drbd_recv_short(*sock, tb, 4, MSG_DONTWAIT | MSG_PEEK);
824 if (rr > 0 || rr == -EAGAIN) {
832 /* Gets called if a connection is established, or if a new minor gets created
834 int drbd_connected(struct drbd_device *device)
838 atomic_set(&device->packet_seq, 0);
839 device->peer_seq = 0;
841 device->state_mutex = first_peer_device(device)->connection->agreed_pro_version < 100 ?
842 &first_peer_device(device)->connection->cstate_mutex :
843 &device->own_state_mutex;
845 err = drbd_send_sync_param(device);
847 err = drbd_send_sizes(device, 0, 0);
849 err = drbd_send_uuids(device);
851 err = drbd_send_current_state(device);
852 clear_bit(USE_DEGR_WFC_T, &device->flags);
853 clear_bit(RESIZE_PENDING, &device->flags);
854 atomic_set(&device->ap_in_flight, 0);
855 mod_timer(&device->request_timer, jiffies + HZ); /* just start it here. */
861 * 1 yes, we have a valid connection
862 * 0 oops, did not work out, please try again
863 * -1 peer talks different language,
864 * no point in trying again, please go standalone.
865 * -2 We do not have a network config...
867 static int conn_connect(struct drbd_connection *connection)
869 struct drbd_socket sock, msock;
870 struct drbd_device *device;
872 int vnr, timeout, h, ok;
873 bool discard_my_data;
874 enum drbd_state_rv rv;
875 struct accept_wait_data ad = {
876 .connection = connection,
877 .door_bell = COMPLETION_INITIALIZER_ONSTACK(ad.door_bell),
880 clear_bit(DISCONNECT_SENT, &connection->flags);
881 if (conn_request_state(connection, NS(conn, C_WF_CONNECTION), CS_VERBOSE) < SS_SUCCESS)
884 mutex_init(&sock.mutex);
885 sock.sbuf = connection->data.sbuf;
886 sock.rbuf = connection->data.rbuf;
888 mutex_init(&msock.mutex);
889 msock.sbuf = connection->meta.sbuf;
890 msock.rbuf = connection->meta.rbuf;
893 /* Assume that the peer only understands protocol 80 until we know better. */
894 connection->agreed_pro_version = 80;
896 if (prepare_listen_socket(connection, &ad))
902 s = drbd_try_connect(connection);
906 send_first_packet(connection, &sock, P_INITIAL_DATA);
907 } else if (!msock.socket) {
908 clear_bit(RESOLVE_CONFLICTS, &connection->flags);
910 send_first_packet(connection, &msock, P_INITIAL_META);
912 conn_err(connection, "Logic error in conn_connect()\n");
913 goto out_release_sockets;
917 if (sock.socket && msock.socket) {
919 nc = rcu_dereference(connection->net_conf);
920 timeout = nc->ping_timeo * HZ / 10;
922 schedule_timeout_interruptible(timeout);
923 ok = drbd_socket_okay(&sock.socket);
924 ok = drbd_socket_okay(&msock.socket) && ok;
930 s = drbd_wait_for_connect(connection, &ad);
932 int fp = receive_first_packet(connection, s);
933 drbd_socket_okay(&sock.socket);
934 drbd_socket_okay(&msock.socket);
938 conn_warn(connection, "initial packet S crossed\n");
939 sock_release(sock.socket);
946 set_bit(RESOLVE_CONFLICTS, &connection->flags);
948 conn_warn(connection, "initial packet M crossed\n");
949 sock_release(msock.socket);
956 conn_warn(connection, "Error receiving initial packet\n");
959 if (prandom_u32() & 1)
964 if (connection->cstate <= C_DISCONNECTING)
965 goto out_release_sockets;
966 if (signal_pending(current)) {
967 flush_signals(current);
969 if (get_t_state(&connection->receiver) == EXITING)
970 goto out_release_sockets;
973 ok = drbd_socket_okay(&sock.socket);
974 ok = drbd_socket_okay(&msock.socket) && ok;
978 sock_release(ad.s_listen);
980 sock.socket->sk->sk_reuse = SK_CAN_REUSE; /* SO_REUSEADDR */
981 msock.socket->sk->sk_reuse = SK_CAN_REUSE; /* SO_REUSEADDR */
983 sock.socket->sk->sk_allocation = GFP_NOIO;
984 msock.socket->sk->sk_allocation = GFP_NOIO;
986 sock.socket->sk->sk_priority = TC_PRIO_INTERACTIVE_BULK;
987 msock.socket->sk->sk_priority = TC_PRIO_INTERACTIVE;
990 * sock.socket->sk->sk_sndtimeo = connection->net_conf->timeout*HZ/10;
991 * sock.socket->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
992 * first set it to the P_CONNECTION_FEATURES timeout,
993 * which we set to 4x the configured ping_timeout. */
995 nc = rcu_dereference(connection->net_conf);
997 sock.socket->sk->sk_sndtimeo =
998 sock.socket->sk->sk_rcvtimeo = nc->ping_timeo*4*HZ/10;
1000 msock.socket->sk->sk_rcvtimeo = nc->ping_int*HZ;
1001 timeout = nc->timeout * HZ / 10;
1002 discard_my_data = nc->discard_my_data;
1005 msock.socket->sk->sk_sndtimeo = timeout;
1007 /* we don't want delays.
1008 * we use TCP_CORK where appropriate, though */
1009 drbd_tcp_nodelay(sock.socket);
1010 drbd_tcp_nodelay(msock.socket);
1012 connection->data.socket = sock.socket;
1013 connection->meta.socket = msock.socket;
1014 connection->last_received = jiffies;
1016 h = drbd_do_features(connection);
1020 if (connection->cram_hmac_tfm) {
1021 /* drbd_request_state(device, NS(conn, WFAuth)); */
1022 switch (drbd_do_auth(connection)) {
1024 conn_err(connection, "Authentication of peer failed\n");
1027 conn_err(connection, "Authentication of peer failed, trying again.\n");
1032 connection->data.socket->sk->sk_sndtimeo = timeout;
1033 connection->data.socket->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1035 if (drbd_send_protocol(connection) == -EOPNOTSUPP)
1038 set_bit(STATE_SENT, &connection->flags);
1041 idr_for_each_entry(&connection->volumes, device, vnr) {
1042 kref_get(&device->kref);
1045 /* Prevent a race between resync-handshake and
1046 * being promoted to Primary.
1048 * Grab and release the state mutex, so we know that any current
1049 * drbd_set_role() is finished, and any incoming drbd_set_role
1050 * will see the STATE_SENT flag, and wait for it to be cleared.
1052 mutex_lock(device->state_mutex);
1053 mutex_unlock(device->state_mutex);
1055 if (discard_my_data)
1056 set_bit(DISCARD_MY_DATA, &device->flags);
1058 clear_bit(DISCARD_MY_DATA, &device->flags);
1060 drbd_connected(device);
1061 kref_put(&device->kref, drbd_destroy_device);
1066 rv = conn_request_state(connection, NS(conn, C_WF_REPORT_PARAMS), CS_VERBOSE);
1067 if (rv < SS_SUCCESS || connection->cstate != C_WF_REPORT_PARAMS) {
1068 clear_bit(STATE_SENT, &connection->flags);
1072 drbd_thread_start(&connection->asender);
1074 mutex_lock(&connection->conf_update);
1075 /* The discard_my_data flag is a single-shot modifier to the next
1076 * connection attempt, the handshake of which is now well underway.
1077 * No need for rcu style copying of the whole struct
1078 * just to clear a single value. */
1079 connection->net_conf->discard_my_data = 0;
1080 mutex_unlock(&connection->conf_update);
1084 out_release_sockets:
1086 sock_release(ad.s_listen);
1088 sock_release(sock.socket);
1090 sock_release(msock.socket);
1094 static int decode_header(struct drbd_connection *connection, void *header, struct packet_info *pi)
1096 unsigned int header_size = drbd_header_size(connection);
1098 if (header_size == sizeof(struct p_header100) &&
1099 *(__be32 *)header == cpu_to_be32(DRBD_MAGIC_100)) {
1100 struct p_header100 *h = header;
1102 conn_err(connection, "Header padding is not zero\n");
1105 pi->vnr = be16_to_cpu(h->volume);
1106 pi->cmd = be16_to_cpu(h->command);
1107 pi->size = be32_to_cpu(h->length);
1108 } else if (header_size == sizeof(struct p_header95) &&
1109 *(__be16 *)header == cpu_to_be16(DRBD_MAGIC_BIG)) {
1110 struct p_header95 *h = header;
1111 pi->cmd = be16_to_cpu(h->command);
1112 pi->size = be32_to_cpu(h->length);
1114 } else if (header_size == sizeof(struct p_header80) &&
1115 *(__be32 *)header == cpu_to_be32(DRBD_MAGIC)) {
1116 struct p_header80 *h = header;
1117 pi->cmd = be16_to_cpu(h->command);
1118 pi->size = be16_to_cpu(h->length);
1121 conn_err(connection, "Wrong magic value 0x%08x in protocol version %d\n",
1122 be32_to_cpu(*(__be32 *)header),
1123 connection->agreed_pro_version);
1126 pi->data = header + header_size;
1130 static int drbd_recv_header(struct drbd_connection *connection, struct packet_info *pi)
1132 void *buffer = connection->data.rbuf;
1135 err = drbd_recv_all_warn(connection, buffer, drbd_header_size(connection));
1139 err = decode_header(connection, buffer, pi);
1140 connection->last_received = jiffies;
1145 static void drbd_flush(struct drbd_connection *connection)
1148 struct drbd_device *device;
1151 if (connection->write_ordering >= WO_bdev_flush) {
1153 idr_for_each_entry(&connection->volumes, device, vnr) {
1154 if (!get_ldev(device))
1156 kref_get(&device->kref);
1159 rv = blkdev_issue_flush(device->ldev->backing_bdev,
1162 dev_info(DEV, "local disk flush failed with status %d\n", rv);
1163 /* would rather check on EOPNOTSUPP, but that is not reliable.
1164 * don't try again for ANY return value != 0
1165 * if (rv == -EOPNOTSUPP) */
1166 drbd_bump_write_ordering(connection, WO_drain_io);
1169 kref_put(&device->kref, drbd_destroy_device);
1180 * drbd_may_finish_epoch() - Applies an epoch_event to the epoch's state, eventually finishes it.
1181 * @device: DRBD device.
1182 * @epoch: Epoch object.
1185 static enum finish_epoch drbd_may_finish_epoch(struct drbd_connection *connection,
1186 struct drbd_epoch *epoch,
1187 enum epoch_event ev)
1190 struct drbd_epoch *next_epoch;
1191 enum finish_epoch rv = FE_STILL_LIVE;
1193 spin_lock(&connection->epoch_lock);
1197 epoch_size = atomic_read(&epoch->epoch_size);
1199 switch (ev & ~EV_CLEANUP) {
1201 atomic_dec(&epoch->active);
1203 case EV_GOT_BARRIER_NR:
1204 set_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags);
1206 case EV_BECAME_LAST:
1211 if (epoch_size != 0 &&
1212 atomic_read(&epoch->active) == 0 &&
1213 (test_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags) || ev & EV_CLEANUP)) {
1214 if (!(ev & EV_CLEANUP)) {
1215 spin_unlock(&connection->epoch_lock);
1216 drbd_send_b_ack(epoch->connection, epoch->barrier_nr, epoch_size);
1217 spin_lock(&connection->epoch_lock);
1220 /* FIXME: dec unacked on connection, once we have
1221 * something to count pending connection packets in. */
1222 if (test_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags))
1223 dec_unacked(epoch->connection);
1226 if (connection->current_epoch != epoch) {
1227 next_epoch = list_entry(epoch->list.next, struct drbd_epoch, list);
1228 list_del(&epoch->list);
1229 ev = EV_BECAME_LAST | (ev & EV_CLEANUP);
1230 connection->epochs--;
1233 if (rv == FE_STILL_LIVE)
1237 atomic_set(&epoch->epoch_size, 0);
1238 /* atomic_set(&epoch->active, 0); is already zero */
1239 if (rv == FE_STILL_LIVE)
1250 spin_unlock(&connection->epoch_lock);
1256 * drbd_bump_write_ordering() - Fall back to an other write ordering method
1257 * @connection: DRBD connection.
1258 * @wo: Write ordering method to try.
1260 void drbd_bump_write_ordering(struct drbd_connection *connection, enum write_ordering_e wo)
1262 struct disk_conf *dc;
1263 struct drbd_device *device;
1264 enum write_ordering_e pwo;
1266 static char *write_ordering_str[] = {
1268 [WO_drain_io] = "drain",
1269 [WO_bdev_flush] = "flush",
1272 pwo = connection->write_ordering;
1275 idr_for_each_entry(&connection->volumes, device, vnr) {
1276 if (!get_ldev_if_state(device, D_ATTACHING))
1278 dc = rcu_dereference(device->ldev->disk_conf);
1280 if (wo == WO_bdev_flush && !dc->disk_flushes)
1282 if (wo == WO_drain_io && !dc->disk_drain)
1287 connection->write_ordering = wo;
1288 if (pwo != connection->write_ordering || wo == WO_bdev_flush)
1289 conn_info(connection, "Method to ensure write ordering: %s\n", write_ordering_str[connection->write_ordering]);
1293 * drbd_submit_peer_request()
1294 * @device: DRBD device.
1295 * @peer_req: peer request
1296 * @rw: flag field, see bio->bi_rw
1298 * May spread the pages to multiple bios,
1299 * depending on bio_add_page restrictions.
1301 * Returns 0 if all bios have been submitted,
1302 * -ENOMEM if we could not allocate enough bios,
1303 * -ENOSPC (any better suggestion?) if we have not been able to bio_add_page a
1304 * single page to an empty bio (which should never happen and likely indicates
1305 * that the lower level IO stack is in some way broken). This has been observed
1306 * on certain Xen deployments.
1308 /* TODO allocate from our own bio_set. */
1309 int drbd_submit_peer_request(struct drbd_device *device,
1310 struct drbd_peer_request *peer_req,
1311 const unsigned rw, const int fault_type)
1313 struct bio *bios = NULL;
1315 struct page *page = peer_req->pages;
1316 sector_t sector = peer_req->i.sector;
1317 unsigned ds = peer_req->i.size;
1318 unsigned n_bios = 0;
1319 unsigned nr_pages = (ds + PAGE_SIZE -1) >> PAGE_SHIFT;
1322 /* In most cases, we will only need one bio. But in case the lower
1323 * level restrictions happen to be different at this offset on this
1324 * side than those of the sending peer, we may need to submit the
1325 * request in more than one bio.
1327 * Plain bio_alloc is good enough here, this is no DRBD internally
1328 * generated bio, but a bio allocated on behalf of the peer.
1331 bio = bio_alloc(GFP_NOIO, nr_pages);
1333 dev_err(DEV, "submit_ee: Allocation of a bio failed\n");
1336 /* > peer_req->i.sector, unless this is the first bio */
1337 bio->bi_iter.bi_sector = sector;
1338 bio->bi_bdev = device->ldev->backing_bdev;
1340 bio->bi_private = peer_req;
1341 bio->bi_end_io = drbd_peer_request_endio;
1343 bio->bi_next = bios;
1347 page_chain_for_each(page) {
1348 unsigned len = min_t(unsigned, ds, PAGE_SIZE);
1349 if (!bio_add_page(bio, page, len, 0)) {
1350 /* A single page must always be possible!
1351 * But in case it fails anyways,
1352 * we deal with it, and complain (below). */
1353 if (bio->bi_vcnt == 0) {
1355 "bio_add_page failed for len=%u, "
1356 "bi_vcnt=0 (bi_sector=%llu)\n",
1357 len, (uint64_t)bio->bi_iter.bi_sector);
1367 D_ASSERT(page == NULL);
1370 atomic_set(&peer_req->pending_bios, n_bios);
1373 bios = bios->bi_next;
1374 bio->bi_next = NULL;
1376 drbd_generic_make_request(device, fault_type, bio);
1383 bios = bios->bi_next;
1389 static void drbd_remove_epoch_entry_interval(struct drbd_device *device,
1390 struct drbd_peer_request *peer_req)
1392 struct drbd_interval *i = &peer_req->i;
1394 drbd_remove_interval(&device->write_requests, i);
1395 drbd_clear_interval(i);
1397 /* Wake up any processes waiting for this peer request to complete. */
1399 wake_up(&device->misc_wait);
1402 static void conn_wait_active_ee_empty(struct drbd_connection *connection)
1404 struct drbd_device *device;
1408 idr_for_each_entry(&connection->volumes, device, vnr) {
1409 kref_get(&device->kref);
1411 drbd_wait_ee_list_empty(device, &device->active_ee);
1412 kref_put(&device->kref, drbd_destroy_device);
1418 static int receive_Barrier(struct drbd_connection *connection, struct packet_info *pi)
1421 struct p_barrier *p = pi->data;
1422 struct drbd_epoch *epoch;
1424 /* FIXME these are unacked on connection,
1425 * not a specific (peer)device.
1427 connection->current_epoch->barrier_nr = p->barrier;
1428 connection->current_epoch->connection = connection;
1429 rv = drbd_may_finish_epoch(connection, connection->current_epoch, EV_GOT_BARRIER_NR);
1431 /* P_BARRIER_ACK may imply that the corresponding extent is dropped from
1432 * the activity log, which means it would not be resynced in case the
1433 * R_PRIMARY crashes now.
1434 * Therefore we must send the barrier_ack after the barrier request was
1436 switch (connection->write_ordering) {
1438 if (rv == FE_RECYCLED)
1441 /* receiver context, in the writeout path of the other node.
1442 * avoid potential distributed deadlock */
1443 epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO);
1447 conn_warn(connection, "Allocation of an epoch failed, slowing down\n");
1452 conn_wait_active_ee_empty(connection);
1453 drbd_flush(connection);
1455 if (atomic_read(&connection->current_epoch->epoch_size)) {
1456 epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO);
1463 conn_err(connection, "Strangeness in connection->write_ordering %d\n", connection->write_ordering);
1468 atomic_set(&epoch->epoch_size, 0);
1469 atomic_set(&epoch->active, 0);
1471 spin_lock(&connection->epoch_lock);
1472 if (atomic_read(&connection->current_epoch->epoch_size)) {
1473 list_add(&epoch->list, &connection->current_epoch->list);
1474 connection->current_epoch = epoch;
1475 connection->epochs++;
1477 /* The current_epoch got recycled while we allocated this one... */
1480 spin_unlock(&connection->epoch_lock);
1485 /* used from receive_RSDataReply (recv_resync_read)
1486 * and from receive_Data */
1487 static struct drbd_peer_request *
1488 read_in_block(struct drbd_device *device, u64 id, sector_t sector,
1489 int data_size) __must_hold(local)
1491 const sector_t capacity = drbd_get_capacity(device->this_bdev);
1492 struct drbd_peer_request *peer_req;
1495 void *dig_in = first_peer_device(device)->connection->int_dig_in;
1496 void *dig_vv = first_peer_device(device)->connection->int_dig_vv;
1497 unsigned long *data;
1500 if (first_peer_device(device)->connection->peer_integrity_tfm) {
1501 dgs = crypto_hash_digestsize(first_peer_device(device)->connection->peer_integrity_tfm);
1503 * FIXME: Receive the incoming digest into the receive buffer
1504 * here, together with its struct p_data?
1506 err = drbd_recv_all_warn(first_peer_device(device)->connection, dig_in, dgs);
1512 if (!expect(IS_ALIGNED(data_size, 512)))
1514 if (!expect(data_size <= DRBD_MAX_BIO_SIZE))
1517 /* even though we trust out peer,
1518 * we sometimes have to double check. */
1519 if (sector + (data_size>>9) > capacity) {
1520 dev_err(DEV, "request from peer beyond end of local disk: "
1521 "capacity: %llus < sector: %llus + size: %u\n",
1522 (unsigned long long)capacity,
1523 (unsigned long long)sector, data_size);
1527 /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD
1528 * "criss-cross" setup, that might cause write-out on some other DRBD,
1529 * which in turn might block on the other node at this very place. */
1530 peer_req = drbd_alloc_peer_req(device, id, sector, data_size, GFP_NOIO);
1538 page = peer_req->pages;
1539 page_chain_for_each(page) {
1540 unsigned len = min_t(int, ds, PAGE_SIZE);
1542 err = drbd_recv_all_warn(first_peer_device(device)->connection, data, len);
1543 if (drbd_insert_fault(device, DRBD_FAULT_RECEIVE)) {
1544 dev_err(DEV, "Fault injection: Corrupting data on receive\n");
1545 data[0] = data[0] ^ (unsigned long)-1;
1549 drbd_free_peer_req(device, peer_req);
1556 drbd_csum_ee(device, first_peer_device(device)->connection->peer_integrity_tfm, peer_req, dig_vv);
1557 if (memcmp(dig_in, dig_vv, dgs)) {
1558 dev_err(DEV, "Digest integrity check FAILED: %llus +%u\n",
1559 (unsigned long long)sector, data_size);
1560 drbd_free_peer_req(device, peer_req);
1564 device->recv_cnt += data_size>>9;
1568 /* drbd_drain_block() just takes a data block
1569 * out of the socket input buffer, and discards it.
1571 static int drbd_drain_block(struct drbd_device *device, int data_size)
1580 page = drbd_alloc_pages(device, 1, 1);
1584 unsigned int len = min_t(int, data_size, PAGE_SIZE);
1586 err = drbd_recv_all_warn(first_peer_device(device)->connection, data, len);
1592 drbd_free_pages(device, page, 0);
1596 static int recv_dless_read(struct drbd_device *device, struct drbd_request *req,
1597 sector_t sector, int data_size)
1599 struct bio_vec bvec;
1600 struct bvec_iter iter;
1602 int dgs, err, expect;
1603 void *dig_in = first_peer_device(device)->connection->int_dig_in;
1604 void *dig_vv = first_peer_device(device)->connection->int_dig_vv;
1607 if (first_peer_device(device)->connection->peer_integrity_tfm) {
1608 dgs = crypto_hash_digestsize(first_peer_device(device)->connection->peer_integrity_tfm);
1609 err = drbd_recv_all_warn(first_peer_device(device)->connection, dig_in, dgs);
1615 /* optimistically update recv_cnt. if receiving fails below,
1616 * we disconnect anyways, and counters will be reset. */
1617 device->recv_cnt += data_size>>9;
1619 bio = req->master_bio;
1620 D_ASSERT(sector == bio->bi_iter.bi_sector);
1622 bio_for_each_segment(bvec, bio, iter) {
1623 void *mapped = kmap(bvec.bv_page) + bvec.bv_offset;
1624 expect = min_t(int, data_size, bvec.bv_len);
1625 err = drbd_recv_all_warn(first_peer_device(device)->connection, mapped, expect);
1626 kunmap(bvec.bv_page);
1629 data_size -= expect;
1633 drbd_csum_bio(device, first_peer_device(device)->connection->peer_integrity_tfm, bio, dig_vv);
1634 if (memcmp(dig_in, dig_vv, dgs)) {
1635 dev_err(DEV, "Digest integrity check FAILED. Broken NICs?\n");
1640 D_ASSERT(data_size == 0);
1645 * e_end_resync_block() is called in asender context via
1646 * drbd_finish_peer_reqs().
1648 static int e_end_resync_block(struct drbd_work *w, int unused)
1650 struct drbd_peer_request *peer_req =
1651 container_of(w, struct drbd_peer_request, w);
1652 struct drbd_device *device = w->device;
1653 sector_t sector = peer_req->i.sector;
1656 D_ASSERT(drbd_interval_empty(&peer_req->i));
1658 if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) {
1659 drbd_set_in_sync(device, sector, peer_req->i.size);
1660 err = drbd_send_ack(device, P_RS_WRITE_ACK, peer_req);
1662 /* Record failure to sync */
1663 drbd_rs_failed_io(device, sector, peer_req->i.size);
1665 err = drbd_send_ack(device, P_NEG_ACK, peer_req);
1667 dec_unacked(device);
1672 static int recv_resync_read(struct drbd_device *device, sector_t sector, int data_size) __releases(local)
1674 struct drbd_peer_request *peer_req;
1676 peer_req = read_in_block(device, ID_SYNCER, sector, data_size);
1680 dec_rs_pending(device);
1682 inc_unacked(device);
1683 /* corresponding dec_unacked() in e_end_resync_block()
1684 * respective _drbd_clear_done_ee */
1686 peer_req->w.cb = e_end_resync_block;
1688 spin_lock_irq(&first_peer_device(device)->connection->req_lock);
1689 list_add(&peer_req->w.list, &device->sync_ee);
1690 spin_unlock_irq(&first_peer_device(device)->connection->req_lock);
1692 atomic_add(data_size >> 9, &device->rs_sect_ev);
1693 if (drbd_submit_peer_request(device, peer_req, WRITE, DRBD_FAULT_RS_WR) == 0)
1696 /* don't care for the reason here */
1697 dev_err(DEV, "submit failed, triggering re-connect\n");
1698 spin_lock_irq(&first_peer_device(device)->connection->req_lock);
1699 list_del(&peer_req->w.list);
1700 spin_unlock_irq(&first_peer_device(device)->connection->req_lock);
1702 drbd_free_peer_req(device, peer_req);
1708 static struct drbd_request *
1709 find_request(struct drbd_device *device, struct rb_root *root, u64 id,
1710 sector_t sector, bool missing_ok, const char *func)
1712 struct drbd_request *req;
1714 /* Request object according to our peer */
1715 req = (struct drbd_request *)(unsigned long)id;
1716 if (drbd_contains_interval(root, sector, &req->i) && req->i.local)
1719 dev_err(DEV, "%s: failed to find request 0x%lx, sector %llus\n", func,
1720 (unsigned long)id, (unsigned long long)sector);
1725 static int receive_DataReply(struct drbd_connection *connection, struct packet_info *pi)
1727 struct drbd_device *device;
1728 struct drbd_request *req;
1731 struct p_data *p = pi->data;
1733 device = vnr_to_device(connection, pi->vnr);
1737 sector = be64_to_cpu(p->sector);
1739 spin_lock_irq(&first_peer_device(device)->connection->req_lock);
1740 req = find_request(device, &device->read_requests, p->block_id, sector, false, __func__);
1741 spin_unlock_irq(&first_peer_device(device)->connection->req_lock);
1745 /* hlist_del(&req->collision) is done in _req_may_be_done, to avoid
1746 * special casing it there for the various failure cases.
1747 * still no race with drbd_fail_pending_reads */
1748 err = recv_dless_read(device, req, sector, pi->size);
1750 req_mod(req, DATA_RECEIVED);
1751 /* else: nothing. handled from drbd_disconnect...
1752 * I don't think we may complete this just yet
1753 * in case we are "on-disconnect: freeze" */
1758 static int receive_RSDataReply(struct drbd_connection *connection, struct packet_info *pi)
1760 struct drbd_device *device;
1763 struct p_data *p = pi->data;
1765 device = vnr_to_device(connection, pi->vnr);
1769 sector = be64_to_cpu(p->sector);
1770 D_ASSERT(p->block_id == ID_SYNCER);
1772 if (get_ldev(device)) {
1773 /* data is submitted to disk within recv_resync_read.
1774 * corresponding put_ldev done below on error,
1775 * or in drbd_peer_request_endio. */
1776 err = recv_resync_read(device, sector, pi->size);
1778 if (__ratelimit(&drbd_ratelimit_state))
1779 dev_err(DEV, "Can not write resync data to local disk.\n");
1781 err = drbd_drain_block(device, pi->size);
1783 drbd_send_ack_dp(device, P_NEG_ACK, p, pi->size);
1786 atomic_add(pi->size >> 9, &device->rs_sect_in);
1791 static void restart_conflicting_writes(struct drbd_device *device,
1792 sector_t sector, int size)
1794 struct drbd_interval *i;
1795 struct drbd_request *req;
1797 drbd_for_each_overlap(i, &device->write_requests, sector, size) {
1800 req = container_of(i, struct drbd_request, i);
1801 if (req->rq_state & RQ_LOCAL_PENDING ||
1802 !(req->rq_state & RQ_POSTPONED))
1804 /* as it is RQ_POSTPONED, this will cause it to
1805 * be queued on the retry workqueue. */
1806 __req_mod(req, CONFLICT_RESOLVED, NULL);
1811 * e_end_block() is called in asender context via drbd_finish_peer_reqs().
1813 static int e_end_block(struct drbd_work *w, int cancel)
1815 struct drbd_peer_request *peer_req =
1816 container_of(w, struct drbd_peer_request, w);
1817 struct drbd_device *device = w->device;
1818 sector_t sector = peer_req->i.sector;
1821 if (peer_req->flags & EE_SEND_WRITE_ACK) {
1822 if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) {
1823 pcmd = (device->state.conn >= C_SYNC_SOURCE &&
1824 device->state.conn <= C_PAUSED_SYNC_T &&
1825 peer_req->flags & EE_MAY_SET_IN_SYNC) ?
1826 P_RS_WRITE_ACK : P_WRITE_ACK;
1827 err = drbd_send_ack(device, pcmd, peer_req);
1828 if (pcmd == P_RS_WRITE_ACK)
1829 drbd_set_in_sync(device, sector, peer_req->i.size);
1831 err = drbd_send_ack(device, P_NEG_ACK, peer_req);
1832 /* we expect it to be marked out of sync anyways...
1833 * maybe assert this? */
1835 dec_unacked(device);
1837 /* we delete from the conflict detection hash _after_ we sent out the
1838 * P_WRITE_ACK / P_NEG_ACK, to get the sequence number right. */
1839 if (peer_req->flags & EE_IN_INTERVAL_TREE) {
1840 spin_lock_irq(&first_peer_device(device)->connection->req_lock);
1841 D_ASSERT(!drbd_interval_empty(&peer_req->i));
1842 drbd_remove_epoch_entry_interval(device, peer_req);
1843 if (peer_req->flags & EE_RESTART_REQUESTS)
1844 restart_conflicting_writes(device, sector, peer_req->i.size);
1845 spin_unlock_irq(&first_peer_device(device)->connection->req_lock);
1847 D_ASSERT(drbd_interval_empty(&peer_req->i));
1849 drbd_may_finish_epoch(first_peer_device(device)->connection, peer_req->epoch, EV_PUT + (cancel ? EV_CLEANUP : 0));
1854 static int e_send_ack(struct drbd_work *w, enum drbd_packet ack)
1856 struct drbd_device *device = w->device;
1857 struct drbd_peer_request *peer_req =
1858 container_of(w, struct drbd_peer_request, w);
1861 err = drbd_send_ack(device, ack, peer_req);
1862 dec_unacked(device);
1867 static int e_send_superseded(struct drbd_work *w, int unused)
1869 return e_send_ack(w, P_SUPERSEDED);
1872 static int e_send_retry_write(struct drbd_work *w, int unused)
1874 struct drbd_connection *connection = first_peer_device(w->device)->connection;
1876 return e_send_ack(w, connection->agreed_pro_version >= 100 ?
1877 P_RETRY_WRITE : P_SUPERSEDED);
1880 static bool seq_greater(u32 a, u32 b)
1883 * We assume 32-bit wrap-around here.
1884 * For 24-bit wrap-around, we would have to shift:
1887 return (s32)a - (s32)b > 0;
1890 static u32 seq_max(u32 a, u32 b)
1892 return seq_greater(a, b) ? a : b;
1895 static void update_peer_seq(struct drbd_device *device, unsigned int peer_seq)
1897 unsigned int newest_peer_seq;
1899 if (test_bit(RESOLVE_CONFLICTS, &first_peer_device(device)->connection->flags)) {
1900 spin_lock(&device->peer_seq_lock);
1901 newest_peer_seq = seq_max(device->peer_seq, peer_seq);
1902 device->peer_seq = newest_peer_seq;
1903 spin_unlock(&device->peer_seq_lock);
1904 /* wake up only if we actually changed device->peer_seq */
1905 if (peer_seq == newest_peer_seq)
1906 wake_up(&device->seq_wait);
1910 static inline int overlaps(sector_t s1, int l1, sector_t s2, int l2)
1912 return !((s1 + (l1>>9) <= s2) || (s1 >= s2 + (l2>>9)));
1915 /* maybe change sync_ee into interval trees as well? */
1916 static bool overlapping_resync_write(struct drbd_device *device, struct drbd_peer_request *peer_req)
1918 struct drbd_peer_request *rs_req;
1921 spin_lock_irq(&first_peer_device(device)->connection->req_lock);
1922 list_for_each_entry(rs_req, &device->sync_ee, w.list) {
1923 if (overlaps(peer_req->i.sector, peer_req->i.size,
1924 rs_req->i.sector, rs_req->i.size)) {
1929 spin_unlock_irq(&first_peer_device(device)->connection->req_lock);
1934 /* Called from receive_Data.
1935 * Synchronize packets on sock with packets on msock.
1937 * This is here so even when a P_DATA packet traveling via sock overtook an Ack
1938 * packet traveling on msock, they are still processed in the order they have
1941 * Note: we don't care for Ack packets overtaking P_DATA packets.
1943 * In case packet_seq is larger than device->peer_seq number, there are
1944 * outstanding packets on the msock. We wait for them to arrive.
1945 * In case we are the logically next packet, we update device->peer_seq
1946 * ourselves. Correctly handles 32bit wrap around.
1948 * Assume we have a 10 GBit connection, that is about 1<<30 byte per second,
1949 * about 1<<21 sectors per second. So "worst" case, we have 1<<3 == 8 seconds
1950 * for the 24bit wrap (historical atomic_t guarantee on some archs), and we have
1951 * 1<<9 == 512 seconds aka ages for the 32bit wrap around...
1953 * returns 0 if we may process the packet,
1954 * -ERESTARTSYS if we were interrupted (by disconnect signal). */
1955 static int wait_for_and_update_peer_seq(struct drbd_device *device, const u32 peer_seq)
1961 if (!test_bit(RESOLVE_CONFLICTS, &first_peer_device(device)->connection->flags))
1964 spin_lock(&device->peer_seq_lock);
1966 if (!seq_greater(peer_seq - 1, device->peer_seq)) {
1967 device->peer_seq = seq_max(device->peer_seq, peer_seq);
1971 if (signal_pending(current)) {
1977 tp = rcu_dereference(first_peer_device(device)->connection->net_conf)->two_primaries;
1983 /* Only need to wait if two_primaries is enabled */
1984 prepare_to_wait(&device->seq_wait, &wait, TASK_INTERRUPTIBLE);
1985 spin_unlock(&device->peer_seq_lock);
1987 timeout = rcu_dereference(first_peer_device(device)->connection->net_conf)->ping_timeo*HZ/10;
1989 timeout = schedule_timeout(timeout);
1990 spin_lock(&device->peer_seq_lock);
1993 dev_err(DEV, "Timed out waiting for missing ack packets; disconnecting\n");
1997 spin_unlock(&device->peer_seq_lock);
1998 finish_wait(&device->seq_wait, &wait);
2002 /* see also bio_flags_to_wire()
2003 * DRBD_REQ_*, because we need to semantically map the flags to data packet
2004 * flags and back. We may replicate to other kernel versions. */
2005 static unsigned long wire_flags_to_bio(struct drbd_device *device, u32 dpf)
2007 return (dpf & DP_RW_SYNC ? REQ_SYNC : 0) |
2008 (dpf & DP_FUA ? REQ_FUA : 0) |
2009 (dpf & DP_FLUSH ? REQ_FLUSH : 0) |
2010 (dpf & DP_DISCARD ? REQ_DISCARD : 0);
2013 static void fail_postponed_requests(struct drbd_device *device, sector_t sector,
2016 struct drbd_interval *i;
2019 drbd_for_each_overlap(i, &device->write_requests, sector, size) {
2020 struct drbd_request *req;
2021 struct bio_and_error m;
2025 req = container_of(i, struct drbd_request, i);
2026 if (!(req->rq_state & RQ_POSTPONED))
2028 req->rq_state &= ~RQ_POSTPONED;
2029 __req_mod(req, NEG_ACKED, &m);
2030 spin_unlock_irq(&first_peer_device(device)->connection->req_lock);
2032 complete_master_bio(device, &m);
2033 spin_lock_irq(&first_peer_device(device)->connection->req_lock);
2038 static int handle_write_conflicts(struct drbd_device *device,
2039 struct drbd_peer_request *peer_req)
2041 struct drbd_connection *connection = first_peer_device(device)->connection;
2042 bool resolve_conflicts = test_bit(RESOLVE_CONFLICTS, &connection->flags);
2043 sector_t sector = peer_req->i.sector;
2044 const unsigned int size = peer_req->i.size;
2045 struct drbd_interval *i;
2050 * Inserting the peer request into the write_requests tree will prevent
2051 * new conflicting local requests from being added.
2053 drbd_insert_interval(&device->write_requests, &peer_req->i);
2056 drbd_for_each_overlap(i, &device->write_requests, sector, size) {
2057 if (i == &peer_req->i)
2062 * Our peer has sent a conflicting remote request; this
2063 * should not happen in a two-node setup. Wait for the
2064 * earlier peer request to complete.
2066 err = drbd_wait_misc(device, i);
2072 equal = i->sector == sector && i->size == size;
2073 if (resolve_conflicts) {
2075 * If the peer request is fully contained within the
2076 * overlapping request, it can be considered overwritten
2077 * and thus superseded; otherwise, it will be retried
2078 * once all overlapping requests have completed.
2080 bool superseded = i->sector <= sector && i->sector +
2081 (i->size >> 9) >= sector + (size >> 9);
2084 dev_alert(DEV, "Concurrent writes detected: "
2085 "local=%llus +%u, remote=%llus +%u, "
2086 "assuming %s came first\n",
2087 (unsigned long long)i->sector, i->size,
2088 (unsigned long long)sector, size,
2089 superseded ? "local" : "remote");
2091 inc_unacked(device);
2092 peer_req->w.cb = superseded ? e_send_superseded :
2094 list_add_tail(&peer_req->w.list, &device->done_ee);
2095 wake_asender(first_peer_device(device)->connection);
2100 struct drbd_request *req =
2101 container_of(i, struct drbd_request, i);
2104 dev_alert(DEV, "Concurrent writes detected: "
2105 "local=%llus +%u, remote=%llus +%u\n",
2106 (unsigned long long)i->sector, i->size,
2107 (unsigned long long)sector, size);
2109 if (req->rq_state & RQ_LOCAL_PENDING ||
2110 !(req->rq_state & RQ_POSTPONED)) {
2112 * Wait for the node with the discard flag to
2113 * decide if this request has been superseded
2114 * or needs to be retried.
2115 * Requests that have been superseded will
2116 * disappear from the write_requests tree.
2118 * In addition, wait for the conflicting
2119 * request to finish locally before submitting
2120 * the conflicting peer request.
2122 err = drbd_wait_misc(device, &req->i);
2124 _conn_request_state(first_peer_device(device)->connection,
2125 NS(conn, C_TIMEOUT),
2127 fail_postponed_requests(device, sector, size);
2133 * Remember to restart the conflicting requests after
2134 * the new peer request has completed.
2136 peer_req->flags |= EE_RESTART_REQUESTS;
2143 drbd_remove_epoch_entry_interval(device, peer_req);
2147 /* mirrored write */
2148 static int receive_Data(struct drbd_connection *connection, struct packet_info *pi)
2150 struct drbd_device *device;
2152 struct drbd_peer_request *peer_req;
2153 struct p_data *p = pi->data;
2154 u32 peer_seq = be32_to_cpu(p->seq_num);
2159 device = vnr_to_device(connection, pi->vnr);
2163 if (!get_ldev(device)) {
2166 err = wait_for_and_update_peer_seq(device, peer_seq);
2167 drbd_send_ack_dp(device, P_NEG_ACK, p, pi->size);
2168 atomic_inc(&connection->current_epoch->epoch_size);
2169 err2 = drbd_drain_block(device, pi->size);
2176 * Corresponding put_ldev done either below (on various errors), or in
2177 * drbd_peer_request_endio, if we successfully submit the data at the
2178 * end of this function.
2181 sector = be64_to_cpu(p->sector);
2182 peer_req = read_in_block(device, p->block_id, sector, pi->size);
2188 peer_req->w.cb = e_end_block;
2190 dp_flags = be32_to_cpu(p->dp_flags);
2191 rw |= wire_flags_to_bio(device, dp_flags);
2192 if (peer_req->pages == NULL) {
2193 D_ASSERT(peer_req->i.size == 0);
2194 D_ASSERT(dp_flags & DP_FLUSH);
2197 if (dp_flags & DP_MAY_SET_IN_SYNC)
2198 peer_req->flags |= EE_MAY_SET_IN_SYNC;
2200 spin_lock(&connection->epoch_lock);
2201 peer_req->epoch = connection->current_epoch;
2202 atomic_inc(&peer_req->epoch->epoch_size);
2203 atomic_inc(&peer_req->epoch->active);
2204 spin_unlock(&connection->epoch_lock);
2207 tp = rcu_dereference(first_peer_device(device)->connection->net_conf)->two_primaries;
2210 peer_req->flags |= EE_IN_INTERVAL_TREE;
2211 err = wait_for_and_update_peer_seq(device, peer_seq);
2213 goto out_interrupted;
2214 spin_lock_irq(&first_peer_device(device)->connection->req_lock);
2215 err = handle_write_conflicts(device, peer_req);
2217 spin_unlock_irq(&first_peer_device(device)->connection->req_lock);
2218 if (err == -ENOENT) {
2222 goto out_interrupted;
2225 update_peer_seq(device, peer_seq);
2226 spin_lock_irq(&first_peer_device(device)->connection->req_lock);
2228 list_add(&peer_req->w.list, &device->active_ee);
2229 spin_unlock_irq(&first_peer_device(device)->connection->req_lock);
2231 if (device->state.conn == C_SYNC_TARGET)
2232 wait_event(device->ee_wait, !overlapping_resync_write(device, peer_req));
2234 if (first_peer_device(device)->connection->agreed_pro_version < 100) {
2236 switch (rcu_dereference(first_peer_device(device)->connection->net_conf)->wire_protocol) {
2238 dp_flags |= DP_SEND_WRITE_ACK;
2241 dp_flags |= DP_SEND_RECEIVE_ACK;
2247 if (dp_flags & DP_SEND_WRITE_ACK) {
2248 peer_req->flags |= EE_SEND_WRITE_ACK;
2249 inc_unacked(device);
2250 /* corresponding dec_unacked() in e_end_block()
2251 * respective _drbd_clear_done_ee */
2254 if (dp_flags & DP_SEND_RECEIVE_ACK) {
2255 /* I really don't like it that the receiver thread
2256 * sends on the msock, but anyways */
2257 drbd_send_ack(device, P_RECV_ACK, peer_req);
2260 if (device->state.pdsk < D_INCONSISTENT) {
2261 /* In case we have the only disk of the cluster, */
2262 drbd_set_out_of_sync(device, peer_req->i.sector, peer_req->i.size);
2263 peer_req->flags |= EE_CALL_AL_COMPLETE_IO;
2264 peer_req->flags &= ~EE_MAY_SET_IN_SYNC;
2265 drbd_al_begin_io(device, &peer_req->i, true);
2268 err = drbd_submit_peer_request(device, peer_req, rw, DRBD_FAULT_DT_WR);
2272 /* don't care for the reason here */
2273 dev_err(DEV, "submit failed, triggering re-connect\n");
2274 spin_lock_irq(&first_peer_device(device)->connection->req_lock);
2275 list_del(&peer_req->w.list);
2276 drbd_remove_epoch_entry_interval(device, peer_req);
2277 spin_unlock_irq(&first_peer_device(device)->connection->req_lock);
2278 if (peer_req->flags & EE_CALL_AL_COMPLETE_IO)
2279 drbd_al_complete_io(device, &peer_req->i);
2282 drbd_may_finish_epoch(connection, peer_req->epoch, EV_PUT + EV_CLEANUP);
2284 drbd_free_peer_req(device, peer_req);
2288 /* We may throttle resync, if the lower device seems to be busy,
2289 * and current sync rate is above c_min_rate.
2291 * To decide whether or not the lower device is busy, we use a scheme similar
2292 * to MD RAID is_mddev_idle(): if the partition stats reveal "significant"
2293 * (more than 64 sectors) of activity we cannot account for with our own resync
2294 * activity, it obviously is "busy".
2296 * The current sync rate used here uses only the most recent two step marks,
2297 * to have a short time average so we can react faster.
2299 int drbd_rs_should_slow_down(struct drbd_device *device, sector_t sector)
2301 struct gendisk *disk = device->ldev->backing_bdev->bd_contains->bd_disk;
2302 unsigned long db, dt, dbdt;
2303 struct lc_element *tmp;
2306 unsigned int c_min_rate;
2309 c_min_rate = rcu_dereference(device->ldev->disk_conf)->c_min_rate;
2312 /* feature disabled? */
2313 if (c_min_rate == 0)
2316 spin_lock_irq(&device->al_lock);
2317 tmp = lc_find(device->resync, BM_SECT_TO_EXT(sector));
2319 struct bm_extent *bm_ext = lc_entry(tmp, struct bm_extent, lce);
2320 if (test_bit(BME_PRIORITY, &bm_ext->flags)) {
2321 spin_unlock_irq(&device->al_lock);
2324 /* Do not slow down if app IO is already waiting for this extent */
2326 spin_unlock_irq(&device->al_lock);
2328 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
2329 (int)part_stat_read(&disk->part0, sectors[1]) -
2330 atomic_read(&device->rs_sect_ev);
2332 if (!device->rs_last_events || curr_events - device->rs_last_events > 64) {
2333 unsigned long rs_left;
2336 device->rs_last_events = curr_events;
2338 /* sync speed average over the last 2*DRBD_SYNC_MARK_STEP,
2340 i = (device->rs_last_mark + DRBD_SYNC_MARKS-1) % DRBD_SYNC_MARKS;
2342 if (device->state.conn == C_VERIFY_S || device->state.conn == C_VERIFY_T)
2343 rs_left = device->ov_left;
2345 rs_left = drbd_bm_total_weight(device) - device->rs_failed;
2347 dt = ((long)jiffies - (long)device->rs_mark_time[i]) / HZ;
2350 db = device->rs_mark_left[i] - rs_left;
2351 dbdt = Bit2KB(db/dt);
2353 if (dbdt > c_min_rate)
2360 static int receive_DataRequest(struct drbd_connection *connection, struct packet_info *pi)
2362 struct drbd_device *device;
2365 struct drbd_peer_request *peer_req;
2366 struct digest_info *di = NULL;
2368 unsigned int fault_type;
2369 struct p_block_req *p = pi->data;
2371 device = vnr_to_device(connection, pi->vnr);
2374 capacity = drbd_get_capacity(device->this_bdev);
2376 sector = be64_to_cpu(p->sector);
2377 size = be32_to_cpu(p->blksize);
2379 if (size <= 0 || !IS_ALIGNED(size, 512) || size > DRBD_MAX_BIO_SIZE) {
2380 dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__,
2381 (unsigned long long)sector, size);
2384 if (sector + (size>>9) > capacity) {
2385 dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__,
2386 (unsigned long long)sector, size);
2390 if (!get_ldev_if_state(device, D_UP_TO_DATE)) {
2393 case P_DATA_REQUEST:
2394 drbd_send_ack_rp(device, P_NEG_DREPLY, p);
2396 case P_RS_DATA_REQUEST:
2397 case P_CSUM_RS_REQUEST:
2399 drbd_send_ack_rp(device, P_NEG_RS_DREPLY , p);
2403 dec_rs_pending(device);
2404 drbd_send_ack_ex(device, P_OV_RESULT, sector, size, ID_IN_SYNC);
2409 if (verb && __ratelimit(&drbd_ratelimit_state))
2410 dev_err(DEV, "Can not satisfy peer's read request, "
2411 "no local data.\n");
2413 /* drain possibly payload */
2414 return drbd_drain_block(device, pi->size);
2417 /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD
2418 * "criss-cross" setup, that might cause write-out on some other DRBD,
2419 * which in turn might block on the other node at this very place. */
2420 peer_req = drbd_alloc_peer_req(device, p->block_id, sector, size, GFP_NOIO);
2427 case P_DATA_REQUEST:
2428 peer_req->w.cb = w_e_end_data_req;
2429 fault_type = DRBD_FAULT_DT_RD;
2430 /* application IO, don't drbd_rs_begin_io */
2433 case P_RS_DATA_REQUEST:
2434 peer_req->w.cb = w_e_end_rsdata_req;
2435 fault_type = DRBD_FAULT_RS_RD;
2436 /* used in the sector offset progress display */
2437 device->bm_resync_fo = BM_SECT_TO_BIT(sector);
2441 case P_CSUM_RS_REQUEST:
2442 fault_type = DRBD_FAULT_RS_RD;
2443 di = kmalloc(sizeof(*di) + pi->size, GFP_NOIO);
2447 di->digest_size = pi->size;
2448 di->digest = (((char *)di)+sizeof(struct digest_info));
2450 peer_req->digest = di;
2451 peer_req->flags |= EE_HAS_DIGEST;
2453 if (drbd_recv_all(first_peer_device(device)->connection, di->digest, pi->size))
2456 if (pi->cmd == P_CSUM_RS_REQUEST) {
2457 D_ASSERT(first_peer_device(device)->connection->agreed_pro_version >= 89);
2458 peer_req->w.cb = w_e_end_csum_rs_req;
2459 /* used in the sector offset progress display */
2460 device->bm_resync_fo = BM_SECT_TO_BIT(sector);
2461 } else if (pi->cmd == P_OV_REPLY) {
2462 /* track progress, we may need to throttle */
2463 atomic_add(size >> 9, &device->rs_sect_in);
2464 peer_req->w.cb = w_e_end_ov_reply;
2465 dec_rs_pending(device);
2466 /* drbd_rs_begin_io done when we sent this request,
2467 * but accounting still needs to be done. */
2468 goto submit_for_resync;
2473 if (device->ov_start_sector == ~(sector_t)0 &&
2474 first_peer_device(device)->connection->agreed_pro_version >= 90) {
2475 unsigned long now = jiffies;
2477 device->ov_start_sector = sector;
2478 device->ov_position = sector;
2479 device->ov_left = drbd_bm_bits(device) - BM_SECT_TO_BIT(sector);
2480 device->rs_total = device->ov_left;
2481 for (i = 0; i < DRBD_SYNC_MARKS; i++) {
2482 device->rs_mark_left[i] = device->ov_left;
2483 device->rs_mark_time[i] = now;
2485 dev_info(DEV, "Online Verify start sector: %llu\n",
2486 (unsigned long long)sector);
2488 peer_req->w.cb = w_e_end_ov_req;
2489 fault_type = DRBD_FAULT_RS_RD;
2496 /* Throttle, drbd_rs_begin_io and submit should become asynchronous
2497 * wrt the receiver, but it is not as straightforward as it may seem.
2498 * Various places in the resync start and stop logic assume resync
2499 * requests are processed in order, requeuing this on the worker thread
2500 * introduces a bunch of new code for synchronization between threads.
2502 * Unlimited throttling before drbd_rs_begin_io may stall the resync
2503 * "forever", throttling after drbd_rs_begin_io will lock that extent
2504 * for application writes for the same time. For now, just throttle
2505 * here, where the rest of the code expects the receiver to sleep for
2509 /* Throttle before drbd_rs_begin_io, as that locks out application IO;
2510 * this defers syncer requests for some time, before letting at least
2511 * on request through. The resync controller on the receiving side
2512 * will adapt to the incoming rate accordingly.
2514 * We cannot throttle here if remote is Primary/SyncTarget:
2515 * we would also throttle its application reads.
2516 * In that case, throttling is done on the SyncTarget only.
2518 if (device->state.peer != R_PRIMARY && drbd_rs_should_slow_down(device, sector))
2519 schedule_timeout_uninterruptible(HZ/10);
2520 if (drbd_rs_begin_io(device, sector))
2524 atomic_add(size >> 9, &device->rs_sect_ev);
2527 inc_unacked(device);
2528 spin_lock_irq(&first_peer_device(device)->connection->req_lock);
2529 list_add_tail(&peer_req->w.list, &device->read_ee);
2530 spin_unlock_irq(&first_peer_device(device)->connection->req_lock);
2532 if (drbd_submit_peer_request(device, peer_req, READ, fault_type) == 0)
2535 /* don't care for the reason here */
2536 dev_err(DEV, "submit failed, triggering re-connect\n");
2537 spin_lock_irq(&first_peer_device(device)->connection->req_lock);
2538 list_del(&peer_req->w.list);
2539 spin_unlock_irq(&first_peer_device(device)->connection->req_lock);
2540 /* no drbd_rs_complete_io(), we are dropping the connection anyways */
2544 drbd_free_peer_req(device, peer_req);
2548 static int drbd_asb_recover_0p(struct drbd_device *device) __must_hold(local)
2550 int self, peer, rv = -100;
2551 unsigned long ch_self, ch_peer;
2552 enum drbd_after_sb_p after_sb_0p;
2554 self = device->ldev->md.uuid[UI_BITMAP] & 1;
2555 peer = device->p_uuid[UI_BITMAP] & 1;
2557 ch_peer = device->p_uuid[UI_SIZE];
2558 ch_self = device->comm_bm_set;
2561 after_sb_0p = rcu_dereference(first_peer_device(device)->connection->net_conf)->after_sb_0p;
2563 switch (after_sb_0p) {
2565 case ASB_DISCARD_SECONDARY:
2566 case ASB_CALL_HELPER:
2568 dev_err(DEV, "Configuration error.\n");
2570 case ASB_DISCONNECT:
2572 case ASB_DISCARD_YOUNGER_PRI:
2573 if (self == 0 && peer == 1) {
2577 if (self == 1 && peer == 0) {
2581 /* Else fall through to one of the other strategies... */
2582 case ASB_DISCARD_OLDER_PRI:
2583 if (self == 0 && peer == 1) {
2587 if (self == 1 && peer == 0) {
2591 /* Else fall through to one of the other strategies... */
2592 dev_warn(DEV, "Discard younger/older primary did not find a decision\n"
2593 "Using discard-least-changes instead\n");
2594 case ASB_DISCARD_ZERO_CHG:
2595 if (ch_peer == 0 && ch_self == 0) {
2596 rv = test_bit(RESOLVE_CONFLICTS, &first_peer_device(device)->connection->flags)
2600 if (ch_peer == 0) { rv = 1; break; }
2601 if (ch_self == 0) { rv = -1; break; }
2603 if (after_sb_0p == ASB_DISCARD_ZERO_CHG)
2605 case ASB_DISCARD_LEAST_CHG:
2606 if (ch_self < ch_peer)
2608 else if (ch_self > ch_peer)
2610 else /* ( ch_self == ch_peer ) */
2611 /* Well, then use something else. */
2612 rv = test_bit(RESOLVE_CONFLICTS, &first_peer_device(device)->connection->flags)
2615 case ASB_DISCARD_LOCAL:
2618 case ASB_DISCARD_REMOTE:
2625 static int drbd_asb_recover_1p(struct drbd_device *device) __must_hold(local)
2628 enum drbd_after_sb_p after_sb_1p;
2631 after_sb_1p = rcu_dereference(first_peer_device(device)->connection->net_conf)->after_sb_1p;
2633 switch (after_sb_1p) {
2634 case ASB_DISCARD_YOUNGER_PRI:
2635 case ASB_DISCARD_OLDER_PRI:
2636 case ASB_DISCARD_LEAST_CHG:
2637 case ASB_DISCARD_LOCAL:
2638 case ASB_DISCARD_REMOTE:
2639 case ASB_DISCARD_ZERO_CHG:
2640 dev_err(DEV, "Configuration error.\n");
2642 case ASB_DISCONNECT:
2645 hg = drbd_asb_recover_0p(device);
2646 if (hg == -1 && device->state.role == R_SECONDARY)
2648 if (hg == 1 && device->state.role == R_PRIMARY)
2652 rv = drbd_asb_recover_0p(device);
2654 case ASB_DISCARD_SECONDARY:
2655 return device->state.role == R_PRIMARY ? 1 : -1;
2656 case ASB_CALL_HELPER:
2657 hg = drbd_asb_recover_0p(device);
2658 if (hg == -1 && device->state.role == R_PRIMARY) {
2659 enum drbd_state_rv rv2;
2661 /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE,
2662 * we might be here in C_WF_REPORT_PARAMS which is transient.
2663 * we do not need to wait for the after state change work either. */
2664 rv2 = drbd_change_state(device, CS_VERBOSE, NS(role, R_SECONDARY));
2665 if (rv2 != SS_SUCCESS) {
2666 drbd_khelper(device, "pri-lost-after-sb");
2668 dev_warn(DEV, "Successfully gave up primary role.\n");
2678 static int drbd_asb_recover_2p(struct drbd_device *device) __must_hold(local)
2681 enum drbd_after_sb_p after_sb_2p;
2684 after_sb_2p = rcu_dereference(first_peer_device(device)->connection->net_conf)->after_sb_2p;
2686 switch (after_sb_2p) {
2687 case ASB_DISCARD_YOUNGER_PRI:
2688 case ASB_DISCARD_OLDER_PRI:
2689 case ASB_DISCARD_LEAST_CHG:
2690 case ASB_DISCARD_LOCAL:
2691 case ASB_DISCARD_REMOTE:
2693 case ASB_DISCARD_SECONDARY:
2694 case ASB_DISCARD_ZERO_CHG:
2695 dev_err(DEV, "Configuration error.\n");
2698 rv = drbd_asb_recover_0p(device);
2700 case ASB_DISCONNECT:
2702 case ASB_CALL_HELPER:
2703 hg = drbd_asb_recover_0p(device);
2705 enum drbd_state_rv rv2;
2707 /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE,
2708 * we might be here in C_WF_REPORT_PARAMS which is transient.
2709 * we do not need to wait for the after state change work either. */
2710 rv2 = drbd_change_state(device, CS_VERBOSE, NS(role, R_SECONDARY));
2711 if (rv2 != SS_SUCCESS) {
2712 drbd_khelper(device, "pri-lost-after-sb");
2714 dev_warn(DEV, "Successfully gave up primary role.\n");
2724 static void drbd_uuid_dump(struct drbd_device *device, char *text, u64 *uuid,
2725 u64 bits, u64 flags)
2728 dev_info(DEV, "%s uuid info vanished while I was looking!\n", text);
2731 dev_info(DEV, "%s %016llX:%016llX:%016llX:%016llX bits:%llu flags:%llX\n",
2733 (unsigned long long)uuid[UI_CURRENT],
2734 (unsigned long long)uuid[UI_BITMAP],
2735 (unsigned long long)uuid[UI_HISTORY_START],
2736 (unsigned long long)uuid[UI_HISTORY_END],
2737 (unsigned long long)bits,
2738 (unsigned long long)flags);
2742 100 after split brain try auto recover
2743 2 C_SYNC_SOURCE set BitMap
2744 1 C_SYNC_SOURCE use BitMap
2746 -1 C_SYNC_TARGET use BitMap
2747 -2 C_SYNC_TARGET set BitMap
2748 -100 after split brain, disconnect
2749 -1000 unrelated data
2750 -1091 requires proto 91
2751 -1096 requires proto 96
2753 static int drbd_uuid_compare(struct drbd_device *device, int *rule_nr) __must_hold(local)
2758 self = device->ldev->md.uuid[UI_CURRENT] & ~((u64)1);
2759 peer = device->p_uuid[UI_CURRENT] & ~((u64)1);
2762 if (self == UUID_JUST_CREATED && peer == UUID_JUST_CREATED)
2766 if ((self == UUID_JUST_CREATED || self == (u64)0) &&
2767 peer != UUID_JUST_CREATED)
2771 if (self != UUID_JUST_CREATED &&
2772 (peer == UUID_JUST_CREATED || peer == (u64)0))
2776 int rct, dc; /* roles at crash time */
2778 if (device->p_uuid[UI_BITMAP] == (u64)0 && device->ldev->md.uuid[UI_BITMAP] != (u64)0) {
2780 if (first_peer_device(device)->connection->agreed_pro_version < 91)
2783 if ((device->ldev->md.uuid[UI_BITMAP] & ~((u64)1)) == (device->p_uuid[UI_HISTORY_START] & ~((u64)1)) &&
2784 (device->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (device->p_uuid[UI_HISTORY_START + 1] & ~((u64)1))) {
2785 dev_info(DEV, "was SyncSource, missed the resync finished event, corrected myself:\n");
2786 drbd_uuid_move_history(device);
2787 device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[UI_BITMAP];
2788 device->ldev->md.uuid[UI_BITMAP] = 0;
2790 drbd_uuid_dump(device, "self", device->ldev->md.uuid,
2791 device->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(device) : 0, 0);
2794 dev_info(DEV, "was SyncSource (peer failed to write sync_uuid)\n");
2801 if (device->ldev->md.uuid[UI_BITMAP] == (u64)0 && device->p_uuid[UI_BITMAP] != (u64)0) {
2803 if (first_peer_device(device)->connection->agreed_pro_version < 91)
2806 if ((device->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (device->p_uuid[UI_BITMAP] & ~((u64)1)) &&
2807 (device->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == (device->p_uuid[UI_HISTORY_START] & ~((u64)1))) {
2808 dev_info(DEV, "was SyncTarget, peer missed the resync finished event, corrected peer:\n");
2810 device->p_uuid[UI_HISTORY_START + 1] = device->p_uuid[UI_HISTORY_START];
2811 device->p_uuid[UI_HISTORY_START] = device->p_uuid[UI_BITMAP];
2812 device->p_uuid[UI_BITMAP] = 0UL;
2814 drbd_uuid_dump(device, "peer", device->p_uuid, device->p_uuid[UI_SIZE], device->p_uuid[UI_FLAGS]);
2817 dev_info(DEV, "was SyncTarget (failed to write sync_uuid)\n");
2824 /* Common power [off|failure] */
2825 rct = (test_bit(CRASHED_PRIMARY, &device->flags) ? 1 : 0) +
2826 (device->p_uuid[UI_FLAGS] & 2);
2827 /* lowest bit is set when we were primary,
2828 * next bit (weight 2) is set when peer was primary */
2832 case 0: /* !self_pri && !peer_pri */ return 0;
2833 case 1: /* self_pri && !peer_pri */ return 1;
2834 case 2: /* !self_pri && peer_pri */ return -1;
2835 case 3: /* self_pri && peer_pri */
2836 dc = test_bit(RESOLVE_CONFLICTS, &first_peer_device(device)->connection->flags);
2842 peer = device->p_uuid[UI_BITMAP] & ~((u64)1);
2847 peer = device->p_uuid[UI_HISTORY_START] & ~((u64)1);
2849 if (first_peer_device(device)->connection->agreed_pro_version < 96 ?
2850 (device->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) ==
2851 (device->p_uuid[UI_HISTORY_START + 1] & ~((u64)1)) :
2852 peer + UUID_NEW_BM_OFFSET == (device->p_uuid[UI_BITMAP] & ~((u64)1))) {
2853 /* The last P_SYNC_UUID did not get though. Undo the last start of
2854 resync as sync source modifications of the peer's UUIDs. */
2856 if (first_peer_device(device)->connection->agreed_pro_version < 91)
2859 device->p_uuid[UI_BITMAP] = device->p_uuid[UI_HISTORY_START];
2860 device->p_uuid[UI_HISTORY_START] = device->p_uuid[UI_HISTORY_START + 1];
2862 dev_info(DEV, "Lost last syncUUID packet, corrected:\n");
2863 drbd_uuid_dump(device, "peer", device->p_uuid, device->p_uuid[UI_SIZE], device->p_uuid[UI_FLAGS]);
2870 self = device->ldev->md.uuid[UI_CURRENT] & ~((u64)1);
2871 for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) {
2872 peer = device->p_uuid[i] & ~((u64)1);
2878 self = device->ldev->md.uuid[UI_BITMAP] & ~((u64)1);
2879 peer = device->p_uuid[UI_CURRENT] & ~((u64)1);
2884 self = device->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1);
2886 if (first_peer_device(device)->connection->agreed_pro_version < 96 ?
2887 (device->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) ==
2888 (device->p_uuid[UI_HISTORY_START] & ~((u64)1)) :
2889 self + UUID_NEW_BM_OFFSET == (device->ldev->md.uuid[UI_BITMAP] & ~((u64)1))) {
2890 /* The last P_SYNC_UUID did not get though. Undo the last start of
2891 resync as sync source modifications of our UUIDs. */
2893 if (first_peer_device(device)->connection->agreed_pro_version < 91)
2896 __drbd_uuid_set(device, UI_BITMAP, device->ldev->md.uuid[UI_HISTORY_START]);
2897 __drbd_uuid_set(device, UI_HISTORY_START, device->ldev->md.uuid[UI_HISTORY_START + 1]);
2899 dev_info(DEV, "Last syncUUID did not get through, corrected:\n");
2900 drbd_uuid_dump(device, "self", device->ldev->md.uuid,
2901 device->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(device) : 0, 0);
2909 peer = device->p_uuid[UI_CURRENT] & ~((u64)1);
2910 for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) {
2911 self = device->ldev->md.uuid[i] & ~((u64)1);
2917 self = device->ldev->md.uuid[UI_BITMAP] & ~((u64)1);
2918 peer = device->p_uuid[UI_BITMAP] & ~((u64)1);
2919 if (self == peer && self != ((u64)0))
2923 for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) {
2924 self = device->ldev->md.uuid[i] & ~((u64)1);
2925 for (j = UI_HISTORY_START; j <= UI_HISTORY_END; j++) {
2926 peer = device->p_uuid[j] & ~((u64)1);
2935 /* drbd_sync_handshake() returns the new conn state on success, or
2936 CONN_MASK (-1) on failure.
2938 static enum drbd_conns drbd_sync_handshake(struct drbd_device *device, enum drbd_role peer_role,
2939 enum drbd_disk_state peer_disk) __must_hold(local)
2941 enum drbd_conns rv = C_MASK;
2942 enum drbd_disk_state mydisk;
2943 struct net_conf *nc;
2944 int hg, rule_nr, rr_conflict, tentative;
2946 mydisk = device->state.disk;
2947 if (mydisk == D_NEGOTIATING)
2948 mydisk = device->new_state_tmp.disk;
2950 dev_info(DEV, "drbd_sync_handshake:\n");
2952 spin_lock_irq(&device->ldev->md.uuid_lock);
2953 drbd_uuid_dump(device, "self", device->ldev->md.uuid, device->comm_bm_set, 0);
2954 drbd_uuid_dump(device, "peer", device->p_uuid,
2955 device->p_uuid[UI_SIZE], device->p_uuid[UI_FLAGS]);
2957 hg = drbd_uuid_compare(device, &rule_nr);
2958 spin_unlock_irq(&device->ldev->md.uuid_lock);
2960 dev_info(DEV, "uuid_compare()=%d by rule %d\n", hg, rule_nr);
2963 dev_alert(DEV, "Unrelated data, aborting!\n");
2967 dev_alert(DEV, "To resolve this both sides have to support at least protocol %d\n", -hg - 1000);
2971 if ((mydisk == D_INCONSISTENT && peer_disk > D_INCONSISTENT) ||
2972 (peer_disk == D_INCONSISTENT && mydisk > D_INCONSISTENT)) {
2973 int f = (hg == -100) || abs(hg) == 2;
2974 hg = mydisk > D_INCONSISTENT ? 1 : -1;
2977 dev_info(DEV, "Becoming sync %s due to disk states.\n",
2978 hg > 0 ? "source" : "target");
2982 drbd_khelper(device, "initial-split-brain");
2985 nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
2987 if (hg == 100 || (hg == -100 && nc->always_asbp)) {
2988 int pcount = (device->state.role == R_PRIMARY)
2989 + (peer_role == R_PRIMARY);
2990 int forced = (hg == -100);
2994 hg = drbd_asb_recover_0p(device);
2997 hg = drbd_asb_recover_1p(device);
3000 hg = drbd_asb_recover_2p(device);
3003 if (abs(hg) < 100) {
3004 dev_warn(DEV, "Split-Brain detected, %d primaries, "
3005 "automatically solved. Sync from %s node\n",
3006 pcount, (hg < 0) ? "peer" : "this");
3008 dev_warn(DEV, "Doing a full sync, since"
3009 " UUIDs where ambiguous.\n");
3016 if (test_bit(DISCARD_MY_DATA, &device->flags) && !(device->p_uuid[UI_FLAGS]&1))
3018 if (!test_bit(DISCARD_MY_DATA, &device->flags) && (device->p_uuid[UI_FLAGS]&1))
3022 dev_warn(DEV, "Split-Brain detected, manually solved. "
3023 "Sync from %s node\n",
3024 (hg < 0) ? "peer" : "this");
3026 rr_conflict = nc->rr_conflict;
3027 tentative = nc->tentative;
3031 /* FIXME this log message is not correct if we end up here
3032 * after an attempted attach on a diskless node.
3033 * We just refuse to attach -- well, we drop the "connection"
3034 * to that disk, in a way... */
3035 dev_alert(DEV, "Split-Brain detected but unresolved, dropping connection!\n");
3036 drbd_khelper(device, "split-brain");
3040 if (hg > 0 && mydisk <= D_INCONSISTENT) {
3041 dev_err(DEV, "I shall become SyncSource, but I am inconsistent!\n");
3045 if (hg < 0 && /* by intention we do not use mydisk here. */
3046 device->state.role == R_PRIMARY && device->state.disk >= D_CONSISTENT) {
3047 switch (rr_conflict) {
3048 case ASB_CALL_HELPER:
3049 drbd_khelper(device, "pri-lost");
3051 case ASB_DISCONNECT:
3052 dev_err(DEV, "I shall become SyncTarget, but I am primary!\n");
3055 dev_warn(DEV, "Becoming SyncTarget, violating the stable-data"
3060 if (tentative || test_bit(CONN_DRY_RUN, &first_peer_device(device)->connection->flags)) {
3062 dev_info(DEV, "dry-run connect: No resync, would become Connected immediately.\n");
3064 dev_info(DEV, "dry-run connect: Would become %s, doing a %s resync.",
3065 drbd_conn_str(hg > 0 ? C_SYNC_SOURCE : C_SYNC_TARGET),
3066 abs(hg) >= 2 ? "full" : "bit-map based");
3071 dev_info(DEV, "Writing the whole bitmap, full sync required after drbd_sync_handshake.\n");
3072 if (drbd_bitmap_io(device, &drbd_bmio_set_n_write, "set_n_write from sync_handshake",
3073 BM_LOCKED_SET_ALLOWED))
3077 if (hg > 0) { /* become sync source. */
3079 } else if (hg < 0) { /* become sync target */
3083 if (drbd_bm_total_weight(device)) {
3084 dev_info(DEV, "No resync, but %lu bits in bitmap!\n",
3085 drbd_bm_total_weight(device));
3092 static enum drbd_after_sb_p convert_after_sb(enum drbd_after_sb_p peer)
3094 /* ASB_DISCARD_REMOTE - ASB_DISCARD_LOCAL is valid */
3095 if (peer == ASB_DISCARD_REMOTE)
3096 return ASB_DISCARD_LOCAL;
3098 /* any other things with ASB_DISCARD_REMOTE or ASB_DISCARD_LOCAL are invalid */
3099 if (peer == ASB_DISCARD_LOCAL)
3100 return ASB_DISCARD_REMOTE;
3102 /* everything else is valid if they are equal on both sides. */
3106 static int receive_protocol(struct drbd_connection *connection, struct packet_info *pi)
3108 struct p_protocol *p = pi->data;
3109 enum drbd_after_sb_p p_after_sb_0p, p_after_sb_1p, p_after_sb_2p;
3110 int p_proto, p_discard_my_data, p_two_primaries, cf;
3111 struct net_conf *nc, *old_net_conf, *new_net_conf = NULL;
3112 char integrity_alg[SHARED_SECRET_MAX] = "";
3113 struct crypto_hash *peer_integrity_tfm = NULL;
3114 void *int_dig_in = NULL, *int_dig_vv = NULL;
3116 p_proto = be32_to_cpu(p->protocol);
3117 p_after_sb_0p = be32_to_cpu(p->after_sb_0p);
3118 p_after_sb_1p = be32_to_cpu(p->after_sb_1p);
3119 p_after_sb_2p = be32_to_cpu(p->after_sb_2p);
3120 p_two_primaries = be32_to_cpu(p->two_primaries);
3121 cf = be32_to_cpu(p->conn_flags);
3122 p_discard_my_data = cf & CF_DISCARD_MY_DATA;
3124 if (connection->agreed_pro_version >= 87) {
3127 if (pi->size > sizeof(integrity_alg))
3129 err = drbd_recv_all(connection, integrity_alg, pi->size);
3132 integrity_alg[SHARED_SECRET_MAX - 1] = 0;
3135 if (pi->cmd != P_PROTOCOL_UPDATE) {
3136 clear_bit(CONN_DRY_RUN, &connection->flags);
3138 if (cf & CF_DRY_RUN)
3139 set_bit(CONN_DRY_RUN, &connection->flags);
3142 nc = rcu_dereference(connection->net_conf);
3144 if (p_proto != nc->wire_protocol) {
3145 conn_err(connection, "incompatible %s settings\n", "protocol");
3146 goto disconnect_rcu_unlock;
3149 if (convert_after_sb(p_after_sb_0p) != nc->after_sb_0p) {
3150 conn_err(connection, "incompatible %s settings\n", "after-sb-0pri");
3151 goto disconnect_rcu_unlock;
3154 if (convert_after_sb(p_after_sb_1p) != nc->after_sb_1p) {
3155 conn_err(connection, "incompatible %s settings\n", "after-sb-1pri");
3156 goto disconnect_rcu_unlock;
3159 if (convert_after_sb(p_after_sb_2p) != nc->after_sb_2p) {
3160 conn_err(connection, "incompatible %s settings\n", "after-sb-2pri");
3161 goto disconnect_rcu_unlock;
3164 if (p_discard_my_data && nc->discard_my_data) {
3165 conn_err(connection, "incompatible %s settings\n", "discard-my-data");
3166 goto disconnect_rcu_unlock;
3169 if (p_two_primaries != nc->two_primaries) {
3170 conn_err(connection, "incompatible %s settings\n", "allow-two-primaries");
3171 goto disconnect_rcu_unlock;
3174 if (strcmp(integrity_alg, nc->integrity_alg)) {
3175 conn_err(connection, "incompatible %s settings\n", "data-integrity-alg");
3176 goto disconnect_rcu_unlock;
3182 if (integrity_alg[0]) {
3186 * We can only change the peer data integrity algorithm
3187 * here. Changing our own data integrity algorithm
3188 * requires that we send a P_PROTOCOL_UPDATE packet at
3189 * the same time; otherwise, the peer has no way to
3190 * tell between which packets the algorithm should
3194 peer_integrity_tfm = crypto_alloc_hash(integrity_alg, 0, CRYPTO_ALG_ASYNC);
3195 if (!peer_integrity_tfm) {
3196 conn_err(connection, "peer data-integrity-alg %s not supported\n",
3201 hash_size = crypto_hash_digestsize(peer_integrity_tfm);
3202 int_dig_in = kmalloc(hash_size, GFP_KERNEL);
3203 int_dig_vv = kmalloc(hash_size, GFP_KERNEL);
3204 if (!(int_dig_in && int_dig_vv)) {
3205 conn_err(connection, "Allocation of buffers for data integrity checking failed\n");
3210 new_net_conf = kmalloc(sizeof(struct net_conf), GFP_KERNEL);
3211 if (!new_net_conf) {
3212 conn_err(connection, "Allocation of new net_conf failed\n");
3216 mutex_lock(&connection->data.mutex);
3217 mutex_lock(&connection->conf_update);
3218 old_net_conf = connection->net_conf;
3219 *new_net_conf = *old_net_conf;
3221 new_net_conf->wire_protocol = p_proto;
3222 new_net_conf->after_sb_0p = convert_after_sb(p_after_sb_0p);
3223 new_net_conf->after_sb_1p = convert_after_sb(p_after_sb_1p);
3224 new_net_conf->after_sb_2p = convert_after_sb(p_after_sb_2p);
3225 new_net_conf->two_primaries = p_two_primaries;
3227 rcu_assign_pointer(connection->net_conf, new_net_conf);
3228 mutex_unlock(&connection->conf_update);
3229 mutex_unlock(&connection->data.mutex);
3231 crypto_free_hash(connection->peer_integrity_tfm);
3232 kfree(connection->int_dig_in);
3233 kfree(connection->int_dig_vv);
3234 connection->peer_integrity_tfm = peer_integrity_tfm;
3235 connection->int_dig_in = int_dig_in;
3236 connection->int_dig_vv = int_dig_vv;
3238 if (strcmp(old_net_conf->integrity_alg, integrity_alg))
3239 conn_info(connection, "peer data-integrity-alg: %s\n",
3240 integrity_alg[0] ? integrity_alg : "(none)");
3243 kfree(old_net_conf);
3246 disconnect_rcu_unlock:
3249 crypto_free_hash(peer_integrity_tfm);
3252 conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD);
3257 * input: alg name, feature name
3258 * return: NULL (alg name was "")
3259 * ERR_PTR(error) if something goes wrong
3260 * or the crypto hash ptr, if it worked out ok. */
3262 struct crypto_hash *drbd_crypto_alloc_digest_safe(const struct drbd_device *device,
3263 const char *alg, const char *name)
3265 struct crypto_hash *tfm;
3270 tfm = crypto_alloc_hash(alg, 0, CRYPTO_ALG_ASYNC);
3272 dev_err(DEV, "Can not allocate \"%s\" as %s (reason: %ld)\n",
3273 alg, name, PTR_ERR(tfm));
3279 static int ignore_remaining_packet(struct drbd_connection *connection, struct packet_info *pi)
3281 void *buffer = connection->data.rbuf;
3282 int size = pi->size;
3285 int s = min_t(int, size, DRBD_SOCKET_BUFFER_SIZE);
3286 s = drbd_recv(connection, buffer, s);
3300 * config_unknown_volume - device configuration command for unknown volume
3302 * When a device is added to an existing connection, the node on which the
3303 * device is added first will send configuration commands to its peer but the
3304 * peer will not know about the device yet. It will warn and ignore these
3305 * commands. Once the device is added on the second node, the second node will
3306 * send the same device configuration commands, but in the other direction.
3308 * (We can also end up here if drbd is misconfigured.)
3310 static int config_unknown_volume(struct drbd_connection *connection, struct packet_info *pi)
3312 conn_warn(connection, "%s packet received for volume %u, which is not configured locally\n",
3313 cmdname(pi->cmd), pi->vnr);
3314 return ignore_remaining_packet(connection, pi);
3317 static int receive_SyncParam(struct drbd_connection *connection, struct packet_info *pi)
3319 struct drbd_device *device;
3320 struct p_rs_param_95 *p;
3321 unsigned int header_size, data_size, exp_max_sz;
3322 struct crypto_hash *verify_tfm = NULL;
3323 struct crypto_hash *csums_tfm = NULL;
3324 struct net_conf *old_net_conf, *new_net_conf = NULL;
3325 struct disk_conf *old_disk_conf = NULL, *new_disk_conf = NULL;
3326 const int apv = connection->agreed_pro_version;
3327 struct fifo_buffer *old_plan = NULL, *new_plan = NULL;
3331 device = vnr_to_device(connection, pi->vnr);
3333 return config_unknown_volume(connection, pi);
3335 exp_max_sz = apv <= 87 ? sizeof(struct p_rs_param)
3336 : apv == 88 ? sizeof(struct p_rs_param)
3338 : apv <= 94 ? sizeof(struct p_rs_param_89)
3339 : /* apv >= 95 */ sizeof(struct p_rs_param_95);
3341 if (pi->size > exp_max_sz) {
3342 dev_err(DEV, "SyncParam packet too long: received %u, expected <= %u bytes\n",
3343 pi->size, exp_max_sz);
3348 header_size = sizeof(struct p_rs_param);
3349 data_size = pi->size - header_size;
3350 } else if (apv <= 94) {
3351 header_size = sizeof(struct p_rs_param_89);
3352 data_size = pi->size - header_size;
3353 D_ASSERT(data_size == 0);
3355 header_size = sizeof(struct p_rs_param_95);
3356 data_size = pi->size - header_size;
3357 D_ASSERT(data_size == 0);
3360 /* initialize verify_alg and csums_alg */
3362 memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX);
3364 err = drbd_recv_all(first_peer_device(device)->connection, p, header_size);
3368 mutex_lock(&first_peer_device(device)->connection->conf_update);
3369 old_net_conf = first_peer_device(device)->connection->net_conf;
3370 if (get_ldev(device)) {
3371 new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL);
3372 if (!new_disk_conf) {
3374 mutex_unlock(&first_peer_device(device)->connection->conf_update);
3375 dev_err(DEV, "Allocation of new disk_conf failed\n");
3379 old_disk_conf = device->ldev->disk_conf;
3380 *new_disk_conf = *old_disk_conf;
3382 new_disk_conf->resync_rate = be32_to_cpu(p->resync_rate);
3387 if (data_size > SHARED_SECRET_MAX || data_size == 0) {
3388 dev_err(DEV, "verify-alg of wrong size, "
3389 "peer wants %u, accepting only up to %u byte\n",
3390 data_size, SHARED_SECRET_MAX);
3395 err = drbd_recv_all(first_peer_device(device)->connection, p->verify_alg, data_size);
3398 /* we expect NUL terminated string */
3399 /* but just in case someone tries to be evil */
3400 D_ASSERT(p->verify_alg[data_size-1] == 0);
3401 p->verify_alg[data_size-1] = 0;
3403 } else /* apv >= 89 */ {
3404 /* we still expect NUL terminated strings */
3405 /* but just in case someone tries to be evil */
3406 D_ASSERT(p->verify_alg[SHARED_SECRET_MAX-1] == 0);
3407 D_ASSERT(p->csums_alg[SHARED_SECRET_MAX-1] == 0);
3408 p->verify_alg[SHARED_SECRET_MAX-1] = 0;
3409 p->csums_alg[SHARED_SECRET_MAX-1] = 0;
3412 if (strcmp(old_net_conf->verify_alg, p->verify_alg)) {
3413 if (device->state.conn == C_WF_REPORT_PARAMS) {
3414 dev_err(DEV, "Different verify-alg settings. me=\"%s\" peer=\"%s\"\n",
3415 old_net_conf->verify_alg, p->verify_alg);
3418 verify_tfm = drbd_crypto_alloc_digest_safe(device,
3419 p->verify_alg, "verify-alg");
3420 if (IS_ERR(verify_tfm)) {
3426 if (apv >= 89 && strcmp(old_net_conf->csums_alg, p->csums_alg)) {
3427 if (device->state.conn == C_WF_REPORT_PARAMS) {
3428 dev_err(DEV, "Different csums-alg settings. me=\"%s\" peer=\"%s\"\n",
3429 old_net_conf->csums_alg, p->csums_alg);
3432 csums_tfm = drbd_crypto_alloc_digest_safe(device,
3433 p->csums_alg, "csums-alg");
3434 if (IS_ERR(csums_tfm)) {
3440 if (apv > 94 && new_disk_conf) {
3441 new_disk_conf->c_plan_ahead = be32_to_cpu(p->c_plan_ahead);
3442 new_disk_conf->c_delay_target = be32_to_cpu(p->c_delay_target);
3443 new_disk_conf->c_fill_target = be32_to_cpu(p->c_fill_target);
3444 new_disk_conf->c_max_rate = be32_to_cpu(p->c_max_rate);
3446 fifo_size = (new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ;
3447 if (fifo_size != device->rs_plan_s->size) {
3448 new_plan = fifo_alloc(fifo_size);
3450 dev_err(DEV, "kmalloc of fifo_buffer failed");
3457 if (verify_tfm || csums_tfm) {
3458 new_net_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
3459 if (!new_net_conf) {
3460 dev_err(DEV, "Allocation of new net_conf failed\n");
3464 *new_net_conf = *old_net_conf;
3467 strcpy(new_net_conf->verify_alg, p->verify_alg);
3468 new_net_conf->verify_alg_len = strlen(p->verify_alg) + 1;
3469 crypto_free_hash(first_peer_device(device)->connection->verify_tfm);
3470 first_peer_device(device)->connection->verify_tfm = verify_tfm;
3471 dev_info(DEV, "using verify-alg: \"%s\"\n", p->verify_alg);
3474 strcpy(new_net_conf->csums_alg, p->csums_alg);
3475 new_net_conf->csums_alg_len = strlen(p->csums_alg) + 1;
3476 crypto_free_hash(first_peer_device(device)->connection->csums_tfm);
3477 first_peer_device(device)->connection->csums_tfm = csums_tfm;
3478 dev_info(DEV, "using csums-alg: \"%s\"\n", p->csums_alg);
3480 rcu_assign_pointer(connection->net_conf, new_net_conf);
3484 if (new_disk_conf) {
3485 rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf);
3490 old_plan = device->rs_plan_s;
3491 rcu_assign_pointer(device->rs_plan_s, new_plan);
3494 mutex_unlock(&first_peer_device(device)->connection->conf_update);
3497 kfree(old_net_conf);
3498 kfree(old_disk_conf);
3504 if (new_disk_conf) {
3506 kfree(new_disk_conf);
3508 mutex_unlock(&first_peer_device(device)->connection->conf_update);
3513 if (new_disk_conf) {
3515 kfree(new_disk_conf);
3517 mutex_unlock(&first_peer_device(device)->connection->conf_update);
3518 /* just for completeness: actually not needed,
3519 * as this is not reached if csums_tfm was ok. */
3520 crypto_free_hash(csums_tfm);
3521 /* but free the verify_tfm again, if csums_tfm did not work out */
3522 crypto_free_hash(verify_tfm);
3523 conn_request_state(first_peer_device(device)->connection, NS(conn, C_DISCONNECTING), CS_HARD);
3527 /* warn if the arguments differ by more than 12.5% */
3528 static void warn_if_differ_considerably(struct drbd_device *device,
3529 const char *s, sector_t a, sector_t b)
3532 if (a == 0 || b == 0)
3534 d = (a > b) ? (a - b) : (b - a);
3535 if (d > (a>>3) || d > (b>>3))
3536 dev_warn(DEV, "Considerable difference in %s: %llus vs. %llus\n", s,
3537 (unsigned long long)a, (unsigned long long)b);
3540 static int receive_sizes(struct drbd_connection *connection, struct packet_info *pi)
3542 struct drbd_device *device;
3543 struct p_sizes *p = pi->data;
3544 enum determine_dev_size dd = DS_UNCHANGED;
3545 sector_t p_size, p_usize, my_usize;
3546 int ldsc = 0; /* local disk size changed */
3547 enum dds_flags ddsf;
3549 device = vnr_to_device(connection, pi->vnr);
3551 return config_unknown_volume(connection, pi);
3553 p_size = be64_to_cpu(p->d_size);
3554 p_usize = be64_to_cpu(p->u_size);
3556 /* just store the peer's disk size for now.
3557 * we still need to figure out whether we accept that. */
3558 device->p_size = p_size;
3560 if (get_ldev(device)) {
3562 my_usize = rcu_dereference(device->ldev->disk_conf)->disk_size;
3565 warn_if_differ_considerably(device, "lower level device sizes",
3566 p_size, drbd_get_max_capacity(device->ldev));
3567 warn_if_differ_considerably(device, "user requested size",
3570 /* if this is the first connect, or an otherwise expected
3571 * param exchange, choose the minimum */
3572 if (device->state.conn == C_WF_REPORT_PARAMS)
3573 p_usize = min_not_zero(my_usize, p_usize);
3575 /* Never shrink a device with usable data during connect.
3576 But allow online shrinking if we are connected. */
3577 if (drbd_new_dev_size(device, device->ldev, p_usize, 0) <
3578 drbd_get_capacity(device->this_bdev) &&
3579 device->state.disk >= D_OUTDATED &&
3580 device->state.conn < C_CONNECTED) {
3581 dev_err(DEV, "The peer's disk size is too small!\n");
3582 conn_request_state(first_peer_device(device)->connection, NS(conn, C_DISCONNECTING), CS_HARD);
3587 if (my_usize != p_usize) {
3588 struct disk_conf *old_disk_conf, *new_disk_conf = NULL;
3590 new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL);
3591 if (!new_disk_conf) {
3592 dev_err(DEV, "Allocation of new disk_conf failed\n");
3597 mutex_lock(&first_peer_device(device)->connection->conf_update);
3598 old_disk_conf = device->ldev->disk_conf;
3599 *new_disk_conf = *old_disk_conf;
3600 new_disk_conf->disk_size = p_usize;
3602 rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf);
3603 mutex_unlock(&first_peer_device(device)->connection->conf_update);
3605 kfree(old_disk_conf);
3607 dev_info(DEV, "Peer sets u_size to %lu sectors\n",
3608 (unsigned long)my_usize);
3614 ddsf = be16_to_cpu(p->dds_flags);
3615 if (get_ldev(device)) {
3616 dd = drbd_determine_dev_size(device, ddsf, NULL);
3620 drbd_md_sync(device);
3622 /* I am diskless, need to accept the peer's size. */
3623 drbd_set_my_capacity(device, p_size);
3626 device->peer_max_bio_size = be32_to_cpu(p->max_bio_size);
3627 drbd_reconsider_max_bio_size(device);
3629 if (get_ldev(device)) {
3630 if (device->ldev->known_size != drbd_get_capacity(device->ldev->backing_bdev)) {
3631 device->ldev->known_size = drbd_get_capacity(device->ldev->backing_bdev);
3638 if (device->state.conn > C_WF_REPORT_PARAMS) {
3639 if (be64_to_cpu(p->c_size) !=
3640 drbd_get_capacity(device->this_bdev) || ldsc) {
3641 /* we have different sizes, probably peer
3642 * needs to know my new size... */
3643 drbd_send_sizes(device, 0, ddsf);
3645 if (test_and_clear_bit(RESIZE_PENDING, &device->flags) ||
3646 (dd == DS_GREW && device->state.conn == C_CONNECTED)) {
3647 if (device->state.pdsk >= D_INCONSISTENT &&
3648 device->state.disk >= D_INCONSISTENT) {
3649 if (ddsf & DDSF_NO_RESYNC)
3650 dev_info(DEV, "Resync of new storage suppressed with --assume-clean\n");
3652 resync_after_online_grow(device);
3654 set_bit(RESYNC_AFTER_NEG, &device->flags);
3661 static int receive_uuids(struct drbd_connection *connection, struct packet_info *pi)
3663 struct drbd_device *device;
3664 struct p_uuids *p = pi->data;
3666 int i, updated_uuids = 0;
3668 device = vnr_to_device(connection, pi->vnr);
3670 return config_unknown_volume(connection, pi);
3672 p_uuid = kmalloc(sizeof(u64)*UI_EXTENDED_SIZE, GFP_NOIO);
3674 dev_err(DEV, "kmalloc of p_uuid failed\n");
3678 for (i = UI_CURRENT; i < UI_EXTENDED_SIZE; i++)
3679 p_uuid[i] = be64_to_cpu(p->uuid[i]);
3681 kfree(device->p_uuid);
3682 device->p_uuid = p_uuid;
3684 if (device->state.conn < C_CONNECTED &&
3685 device->state.disk < D_INCONSISTENT &&
3686 device->state.role == R_PRIMARY &&
3687 (device->ed_uuid & ~((u64)1)) != (p_uuid[UI_CURRENT] & ~((u64)1))) {
3688 dev_err(DEV, "Can only connect to data with current UUID=%016llX\n",
3689 (unsigned long long)device->ed_uuid);
3690 conn_request_state(first_peer_device(device)->connection, NS(conn, C_DISCONNECTING), CS_HARD);
3694 if (get_ldev(device)) {
3695 int skip_initial_sync =
3696 device->state.conn == C_CONNECTED &&
3697 first_peer_device(device)->connection->agreed_pro_version >= 90 &&
3698 device->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED &&
3699 (p_uuid[UI_FLAGS] & 8);
3700 if (skip_initial_sync) {
3701 dev_info(DEV, "Accepted new current UUID, preparing to skip initial sync\n");
3702 drbd_bitmap_io(device, &drbd_bmio_clear_n_write,
3703 "clear_n_write from receive_uuids",
3704 BM_LOCKED_TEST_ALLOWED);
3705 _drbd_uuid_set(device, UI_CURRENT, p_uuid[UI_CURRENT]);
3706 _drbd_uuid_set(device, UI_BITMAP, 0);
3707 _drbd_set_state(_NS2(device, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
3709 drbd_md_sync(device);
3713 } else if (device->state.disk < D_INCONSISTENT &&
3714 device->state.role == R_PRIMARY) {
3715 /* I am a diskless primary, the peer just created a new current UUID
3717 updated_uuids = drbd_set_ed_uuid(device, p_uuid[UI_CURRENT]);
3720 /* Before we test for the disk state, we should wait until an eventually
3721 ongoing cluster wide state change is finished. That is important if
3722 we are primary and are detaching from our disk. We need to see the
3723 new disk state... */
3724 mutex_lock(device->state_mutex);
3725 mutex_unlock(device->state_mutex);
3726 if (device->state.conn >= C_CONNECTED && device->state.disk < D_INCONSISTENT)
3727 updated_uuids |= drbd_set_ed_uuid(device, p_uuid[UI_CURRENT]);
3730 drbd_print_uuids(device, "receiver updated UUIDs to");
3736 * convert_state() - Converts the peer's view of the cluster state to our point of view
3737 * @ps: The state as seen by the peer.
3739 static union drbd_state convert_state(union drbd_state ps)
3741 union drbd_state ms;
3743 static enum drbd_conns c_tab[] = {
3744 [C_WF_REPORT_PARAMS] = C_WF_REPORT_PARAMS,
3745 [C_CONNECTED] = C_CONNECTED,
3747 [C_STARTING_SYNC_S] = C_STARTING_SYNC_T,
3748 [C_STARTING_SYNC_T] = C_STARTING_SYNC_S,
3749 [C_DISCONNECTING] = C_TEAR_DOWN, /* C_NETWORK_FAILURE, */
3750 [C_VERIFY_S] = C_VERIFY_T,
3756 ms.conn = c_tab[ps.conn];
3761 ms.peer_isp = (ps.aftr_isp | ps.user_isp);
3766 static int receive_req_state(struct drbd_connection *connection, struct packet_info *pi)
3768 struct drbd_device *device;
3769 struct p_req_state *p = pi->data;
3770 union drbd_state mask, val;
3771 enum drbd_state_rv rv;
3773 device = vnr_to_device(connection, pi->vnr);
3777 mask.i = be32_to_cpu(p->mask);
3778 val.i = be32_to_cpu(p->val);
3780 if (test_bit(RESOLVE_CONFLICTS, &first_peer_device(device)->connection->flags) &&
3781 mutex_is_locked(device->state_mutex)) {
3782 drbd_send_sr_reply(device, SS_CONCURRENT_ST_CHG);
3786 mask = convert_state(mask);
3787 val = convert_state(val);
3789 rv = drbd_change_state(device, CS_VERBOSE, mask, val);
3790 drbd_send_sr_reply(device, rv);
3792 drbd_md_sync(device);
3797 static int receive_req_conn_state(struct drbd_connection *connection, struct packet_info *pi)
3799 struct p_req_state *p = pi->data;
3800 union drbd_state mask, val;
3801 enum drbd_state_rv rv;
3803 mask.i = be32_to_cpu(p->mask);
3804 val.i = be32_to_cpu(p->val);
3806 if (test_bit(RESOLVE_CONFLICTS, &connection->flags) &&
3807 mutex_is_locked(&connection->cstate_mutex)) {
3808 conn_send_sr_reply(connection, SS_CONCURRENT_ST_CHG);
3812 mask = convert_state(mask);
3813 val = convert_state(val);
3815 rv = conn_request_state(connection, mask, val, CS_VERBOSE | CS_LOCAL_ONLY | CS_IGN_OUTD_FAIL);
3816 conn_send_sr_reply(connection, rv);
3821 static int receive_state(struct drbd_connection *connection, struct packet_info *pi)
3823 struct drbd_device *device;
3824 struct p_state *p = pi->data;
3825 union drbd_state os, ns, peer_state;
3826 enum drbd_disk_state real_peer_disk;
3827 enum chg_state_flags cs_flags;
3830 device = vnr_to_device(connection, pi->vnr);
3832 return config_unknown_volume(connection, pi);
3834 peer_state.i = be32_to_cpu(p->state);
3836 real_peer_disk = peer_state.disk;
3837 if (peer_state.disk == D_NEGOTIATING) {
3838 real_peer_disk = device->p_uuid[UI_FLAGS] & 4 ? D_INCONSISTENT : D_CONSISTENT;
3839 dev_info(DEV, "real peer disk state = %s\n", drbd_disk_str(real_peer_disk));
3842 spin_lock_irq(&first_peer_device(device)->connection->req_lock);
3844 os = ns = drbd_read_state(device);
3845 spin_unlock_irq(&first_peer_device(device)->connection->req_lock);
3847 /* If some other part of the code (asender thread, timeout)
3848 * already decided to close the connection again,
3849 * we must not "re-establish" it here. */
3850 if (os.conn <= C_TEAR_DOWN)
3853 /* If this is the "end of sync" confirmation, usually the peer disk
3854 * transitions from D_INCONSISTENT to D_UP_TO_DATE. For empty (0 bits
3855 * set) resync started in PausedSyncT, or if the timing of pause-/
3856 * unpause-sync events has been "just right", the peer disk may
3857 * transition from D_CONSISTENT to D_UP_TO_DATE as well.
3859 if ((os.pdsk == D_INCONSISTENT || os.pdsk == D_CONSISTENT) &&
3860 real_peer_disk == D_UP_TO_DATE &&
3861 os.conn > C_CONNECTED && os.disk == D_UP_TO_DATE) {
3862 /* If we are (becoming) SyncSource, but peer is still in sync
3863 * preparation, ignore its uptodate-ness to avoid flapping, it
3864 * will change to inconsistent once the peer reaches active
3866 * It may have changed syncer-paused flags, however, so we
3867 * cannot ignore this completely. */
3868 if (peer_state.conn > C_CONNECTED &&
3869 peer_state.conn < C_SYNC_SOURCE)
3870 real_peer_disk = D_INCONSISTENT;
3872 /* if peer_state changes to connected at the same time,
3873 * it explicitly notifies us that it finished resync.
3874 * Maybe we should finish it up, too? */
3875 else if (os.conn >= C_SYNC_SOURCE &&
3876 peer_state.conn == C_CONNECTED) {
3877 if (drbd_bm_total_weight(device) <= device->rs_failed)
3878 drbd_resync_finished(device);
3883 /* explicit verify finished notification, stop sector reached. */
3884 if (os.conn == C_VERIFY_T && os.disk == D_UP_TO_DATE &&
3885 peer_state.conn == C_CONNECTED && real_peer_disk == D_UP_TO_DATE) {
3886 ov_out_of_sync_print(device);
3887 drbd_resync_finished(device);
3891 /* peer says his disk is inconsistent, while we think it is uptodate,
3892 * and this happens while the peer still thinks we have a sync going on,
3893 * but we think we are already done with the sync.
3894 * We ignore this to avoid flapping pdsk.
3895 * This should not happen, if the peer is a recent version of drbd. */
3896 if (os.pdsk == D_UP_TO_DATE && real_peer_disk == D_INCONSISTENT &&
3897 os.conn == C_CONNECTED && peer_state.conn > C_SYNC_SOURCE)
3898 real_peer_disk = D_UP_TO_DATE;
3900 if (ns.conn == C_WF_REPORT_PARAMS)
3901 ns.conn = C_CONNECTED;
3903 if (peer_state.conn == C_AHEAD)
3906 if (device->p_uuid && peer_state.disk >= D_NEGOTIATING &&
3907 get_ldev_if_state(device, D_NEGOTIATING)) {
3908 int cr; /* consider resync */
3910 /* if we established a new connection */
3911 cr = (os.conn < C_CONNECTED);
3912 /* if we had an established connection
3913 * and one of the nodes newly attaches a disk */
3914 cr |= (os.conn == C_CONNECTED &&
3915 (peer_state.disk == D_NEGOTIATING ||
3916 os.disk == D_NEGOTIATING));
3917 /* if we have both been inconsistent, and the peer has been
3918 * forced to be UpToDate with --overwrite-data */
3919 cr |= test_bit(CONSIDER_RESYNC, &device->flags);
3920 /* if we had been plain connected, and the admin requested to
3921 * start a sync by "invalidate" or "invalidate-remote" */
3922 cr |= (os.conn == C_CONNECTED &&
3923 (peer_state.conn >= C_STARTING_SYNC_S &&
3924 peer_state.conn <= C_WF_BITMAP_T));
3927 ns.conn = drbd_sync_handshake(device, peer_state.role, real_peer_disk);
3930 if (ns.conn == C_MASK) {
3931 ns.conn = C_CONNECTED;
3932 if (device->state.disk == D_NEGOTIATING) {
3933 drbd_force_state(device, NS(disk, D_FAILED));
3934 } else if (peer_state.disk == D_NEGOTIATING) {
3935 dev_err(DEV, "Disk attach process on the peer node was aborted.\n");
3936 peer_state.disk = D_DISKLESS;
3937 real_peer_disk = D_DISKLESS;
3939 if (test_and_clear_bit(CONN_DRY_RUN, &first_peer_device(device)->connection->flags))
3941 D_ASSERT(os.conn == C_WF_REPORT_PARAMS);
3942 conn_request_state(first_peer_device(device)->connection, NS(conn, C_DISCONNECTING), CS_HARD);
3948 spin_lock_irq(&first_peer_device(device)->connection->req_lock);
3949 if (os.i != drbd_read_state(device).i)
3951 clear_bit(CONSIDER_RESYNC, &device->flags);
3952 ns.peer = peer_state.role;
3953 ns.pdsk = real_peer_disk;
3954 ns.peer_isp = (peer_state.aftr_isp | peer_state.user_isp);
3955 if ((ns.conn == C_CONNECTED || ns.conn == C_WF_BITMAP_S) && ns.disk == D_NEGOTIATING)
3956 ns.disk = device->new_state_tmp.disk;
3957 cs_flags = CS_VERBOSE + (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED ? 0 : CS_HARD);
3958 if (ns.pdsk == D_CONSISTENT && drbd_suspended(device) && ns.conn == C_CONNECTED && os.conn < C_CONNECTED &&
3959 test_bit(NEW_CUR_UUID, &device->flags)) {
3960 /* Do not allow tl_restart(RESEND) for a rebooted peer. We can only allow this
3961 for temporal network outages! */
3962 spin_unlock_irq(&first_peer_device(device)->connection->req_lock);
3963 dev_err(DEV, "Aborting Connect, can not thaw IO with an only Consistent peer\n");
3964 tl_clear(first_peer_device(device)->connection);
3965 drbd_uuid_new_current(device);
3966 clear_bit(NEW_CUR_UUID, &device->flags);
3967 conn_request_state(first_peer_device(device)->connection, NS2(conn, C_PROTOCOL_ERROR, susp, 0), CS_HARD);
3970 rv = _drbd_set_state(device, ns, cs_flags, NULL);
3971 ns = drbd_read_state(device);
3972 spin_unlock_irq(&first_peer_device(device)->connection->req_lock);
3974 if (rv < SS_SUCCESS) {
3975 conn_request_state(first_peer_device(device)->connection, NS(conn, C_DISCONNECTING), CS_HARD);
3979 if (os.conn > C_WF_REPORT_PARAMS) {
3980 if (ns.conn > C_CONNECTED && peer_state.conn <= C_CONNECTED &&
3981 peer_state.disk != D_NEGOTIATING ) {
3982 /* we want resync, peer has not yet decided to sync... */
3983 /* Nowadays only used when forcing a node into primary role and
3984 setting its disk to UpToDate with that */
3985 drbd_send_uuids(device);
3986 drbd_send_current_state(device);
3990 clear_bit(DISCARD_MY_DATA, &device->flags);
3992 drbd_md_sync(device); /* update connected indicator, la_size_sect, ... */
3997 static int receive_sync_uuid(struct drbd_connection *connection, struct packet_info *pi)
3999 struct drbd_device *device;
4000 struct p_rs_uuid *p = pi->data;
4002 device = vnr_to_device(connection, pi->vnr);
4006 wait_event(device->misc_wait,
4007 device->state.conn == C_WF_SYNC_UUID ||
4008 device->state.conn == C_BEHIND ||
4009 device->state.conn < C_CONNECTED ||
4010 device->state.disk < D_NEGOTIATING);
4012 /* D_ASSERT( device->state.conn == C_WF_SYNC_UUID ); */
4014 /* Here the _drbd_uuid_ functions are right, current should
4015 _not_ be rotated into the history */
4016 if (get_ldev_if_state(device, D_NEGOTIATING)) {
4017 _drbd_uuid_set(device, UI_CURRENT, be64_to_cpu(p->uuid));
4018 _drbd_uuid_set(device, UI_BITMAP, 0UL);
4020 drbd_print_uuids(device, "updated sync uuid");
4021 drbd_start_resync(device, C_SYNC_TARGET);
4025 dev_err(DEV, "Ignoring SyncUUID packet!\n");
4031 * receive_bitmap_plain
4033 * Return 0 when done, 1 when another iteration is needed, and a negative error
4034 * code upon failure.
4037 receive_bitmap_plain(struct drbd_device *device, unsigned int size,
4038 unsigned long *p, struct bm_xfer_ctx *c)
4040 unsigned int data_size = DRBD_SOCKET_BUFFER_SIZE -
4041 drbd_header_size(first_peer_device(device)->connection);
4042 unsigned int num_words = min_t(size_t, data_size / sizeof(*p),
4043 c->bm_words - c->word_offset);
4044 unsigned int want = num_words * sizeof(*p);
4048 dev_err(DEV, "%s:want (%u) != size (%u)\n", __func__, want, size);
4053 err = drbd_recv_all(first_peer_device(device)->connection, p, want);
4057 drbd_bm_merge_lel(device, c->word_offset, num_words, p);
4059 c->word_offset += num_words;
4060 c->bit_offset = c->word_offset * BITS_PER_LONG;
4061 if (c->bit_offset > c->bm_bits)
4062 c->bit_offset = c->bm_bits;
4067 static enum drbd_bitmap_code dcbp_get_code(struct p_compressed_bm *p)
4069 return (enum drbd_bitmap_code)(p->encoding & 0x0f);
4072 static int dcbp_get_start(struct p_compressed_bm *p)
4074 return (p->encoding & 0x80) != 0;
4077 static int dcbp_get_pad_bits(struct p_compressed_bm *p)
4079 return (p->encoding >> 4) & 0x7;
4085 * Return 0 when done, 1 when another iteration is needed, and a negative error
4086 * code upon failure.
4089 recv_bm_rle_bits(struct drbd_device *device,
4090 struct p_compressed_bm *p,
4091 struct bm_xfer_ctx *c,
4094 struct bitstream bs;
4098 unsigned long s = c->bit_offset;
4100 int toggle = dcbp_get_start(p);
4104 bitstream_init(&bs, p->code, len, dcbp_get_pad_bits(p));
4106 bits = bitstream_get_bits(&bs, &look_ahead, 64);
4110 for (have = bits; have > 0; s += rl, toggle = !toggle) {
4111 bits = vli_decode_bits(&rl, look_ahead);
4117 if (e >= c->bm_bits) {
4118 dev_err(DEV, "bitmap overflow (e:%lu) while decoding bm RLE packet\n", e);
4121 _drbd_bm_set_bits(device, s, e);
4125 dev_err(DEV, "bitmap decoding error: h:%d b:%d la:0x%08llx l:%u/%u\n",
4126 have, bits, look_ahead,
4127 (unsigned int)(bs.cur.b - p->code),
4128 (unsigned int)bs.buf_len);
4131 /* if we consumed all 64 bits, assign 0; >> 64 is "undefined"; */
4132 if (likely(bits < 64))
4133 look_ahead >>= bits;
4138 bits = bitstream_get_bits(&bs, &tmp, 64 - have);
4141 look_ahead |= tmp << have;
4146 bm_xfer_ctx_bit_to_word_offset(c);
4148 return (s != c->bm_bits);
4154 * Return 0 when done, 1 when another iteration is needed, and a negative error
4155 * code upon failure.
4158 decode_bitmap_c(struct drbd_device *device,
4159 struct p_compressed_bm *p,
4160 struct bm_xfer_ctx *c,
4163 if (dcbp_get_code(p) == RLE_VLI_Bits)
4164 return recv_bm_rle_bits(device, p, c, len - sizeof(*p));
4166 /* other variants had been implemented for evaluation,
4167 * but have been dropped as this one turned out to be "best"
4168 * during all our tests. */
4170 dev_err(DEV, "receive_bitmap_c: unknown encoding %u\n", p->encoding);
4171 conn_request_state(first_peer_device(device)->connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD);
4175 void INFO_bm_xfer_stats(struct drbd_device *device,
4176 const char *direction, struct bm_xfer_ctx *c)
4178 /* what would it take to transfer it "plaintext" */
4179 unsigned int header_size = drbd_header_size(first_peer_device(device)->connection);
4180 unsigned int data_size = DRBD_SOCKET_BUFFER_SIZE - header_size;
4181 unsigned int plain =
4182 header_size * (DIV_ROUND_UP(c->bm_words, data_size) + 1) +
4183 c->bm_words * sizeof(unsigned long);
4184 unsigned int total = c->bytes[0] + c->bytes[1];
4187 /* total can not be zero. but just in case: */
4191 /* don't report if not compressed */
4195 /* total < plain. check for overflow, still */
4196 r = (total > UINT_MAX/1000) ? (total / (plain/1000))
4197 : (1000 * total / plain);
4203 dev_info(DEV, "%s bitmap stats [Bytes(packets)]: plain %u(%u), RLE %u(%u), "
4204 "total %u; compression: %u.%u%%\n",
4206 c->bytes[1], c->packets[1],
4207 c->bytes[0], c->packets[0],
4208 total, r/10, r % 10);
4211 /* Since we are processing the bitfield from lower addresses to higher,
4212 it does not matter if the process it in 32 bit chunks or 64 bit
4213 chunks as long as it is little endian. (Understand it as byte stream,
4214 beginning with the lowest byte...) If we would use big endian
4215 we would need to process it from the highest address to the lowest,
4216 in order to be agnostic to the 32 vs 64 bits issue.
4218 returns 0 on failure, 1 if we successfully received it. */
4219 static int receive_bitmap(struct drbd_connection *connection, struct packet_info *pi)
4221 struct drbd_device *device;
4222 struct bm_xfer_ctx c;
4225 device = vnr_to_device(connection, pi->vnr);
4229 drbd_bm_lock(device, "receive bitmap", BM_LOCKED_SET_ALLOWED);
4230 /* you are supposed to send additional out-of-sync information
4231 * if you actually set bits during this phase */
4233 c = (struct bm_xfer_ctx) {
4234 .bm_bits = drbd_bm_bits(device),
4235 .bm_words = drbd_bm_words(device),
4239 if (pi->cmd == P_BITMAP)
4240 err = receive_bitmap_plain(device, pi->size, pi->data, &c);
4241 else if (pi->cmd == P_COMPRESSED_BITMAP) {
4242 /* MAYBE: sanity check that we speak proto >= 90,
4243 * and the feature is enabled! */
4244 struct p_compressed_bm *p = pi->data;
4246 if (pi->size > DRBD_SOCKET_BUFFER_SIZE - drbd_header_size(connection)) {
4247 dev_err(DEV, "ReportCBitmap packet too large\n");
4251 if (pi->size <= sizeof(*p)) {
4252 dev_err(DEV, "ReportCBitmap packet too small (l:%u)\n", pi->size);
4256 err = drbd_recv_all(first_peer_device(device)->connection, p, pi->size);
4259 err = decode_bitmap_c(device, p, &c, pi->size);
4261 dev_warn(DEV, "receive_bitmap: cmd neither ReportBitMap nor ReportCBitMap (is 0x%x)", pi->cmd);
4266 c.packets[pi->cmd == P_BITMAP]++;
4267 c.bytes[pi->cmd == P_BITMAP] += drbd_header_size(connection) + pi->size;
4274 err = drbd_recv_header(first_peer_device(device)->connection, pi);
4279 INFO_bm_xfer_stats(device, "receive", &c);
4281 if (device->state.conn == C_WF_BITMAP_T) {
4282 enum drbd_state_rv rv;
4284 err = drbd_send_bitmap(device);
4287 /* Omit CS_ORDERED with this state transition to avoid deadlocks. */
4288 rv = _drbd_request_state(device, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE);
4289 D_ASSERT(rv == SS_SUCCESS);
4290 } else if (device->state.conn != C_WF_BITMAP_S) {
4291 /* admin may have requested C_DISCONNECTING,
4292 * other threads may have noticed network errors */
4293 dev_info(DEV, "unexpected cstate (%s) in receive_bitmap\n",
4294 drbd_conn_str(device->state.conn));
4299 drbd_bm_unlock(device);
4300 if (!err && device->state.conn == C_WF_BITMAP_S)
4301 drbd_start_resync(device, C_SYNC_SOURCE);
4305 static int receive_skip(struct drbd_connection *connection, struct packet_info *pi)
4307 conn_warn(connection, "skipping unknown optional packet type %d, l: %d!\n",
4310 return ignore_remaining_packet(connection, pi);
4313 static int receive_UnplugRemote(struct drbd_connection *connection, struct packet_info *pi)
4315 /* Make sure we've acked all the TCP data associated
4316 * with the data requests being unplugged */
4317 drbd_tcp_quickack(connection->data.socket);
4322 static int receive_out_of_sync(struct drbd_connection *connection, struct packet_info *pi)
4324 struct drbd_device *device;
4325 struct p_block_desc *p = pi->data;
4327 device = vnr_to_device(connection, pi->vnr);
4331 switch (device->state.conn) {
4332 case C_WF_SYNC_UUID:
4337 dev_err(DEV, "ASSERT FAILED cstate = %s, expected: WFSyncUUID|WFBitMapT|Behind\n",
4338 drbd_conn_str(device->state.conn));
4341 drbd_set_out_of_sync(device, be64_to_cpu(p->sector), be32_to_cpu(p->blksize));
4349 int (*fn)(struct drbd_connection *, struct packet_info *);
4352 static struct data_cmd drbd_cmd_handler[] = {
4353 [P_DATA] = { 1, sizeof(struct p_data), receive_Data },
4354 [P_DATA_REPLY] = { 1, sizeof(struct p_data), receive_DataReply },
4355 [P_RS_DATA_REPLY] = { 1, sizeof(struct p_data), receive_RSDataReply } ,
4356 [P_BARRIER] = { 0, sizeof(struct p_barrier), receive_Barrier } ,
4357 [P_BITMAP] = { 1, 0, receive_bitmap } ,
4358 [P_COMPRESSED_BITMAP] = { 1, 0, receive_bitmap } ,
4359 [P_UNPLUG_REMOTE] = { 0, 0, receive_UnplugRemote },
4360 [P_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest },
4361 [P_RS_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest },
4362 [P_SYNC_PARAM] = { 1, 0, receive_SyncParam },
4363 [P_SYNC_PARAM89] = { 1, 0, receive_SyncParam },
4364 [P_PROTOCOL] = { 1, sizeof(struct p_protocol), receive_protocol },
4365 [P_UUIDS] = { 0, sizeof(struct p_uuids), receive_uuids },
4366 [P_SIZES] = { 0, sizeof(struct p_sizes), receive_sizes },
4367 [P_STATE] = { 0, sizeof(struct p_state), receive_state },
4368 [P_STATE_CHG_REQ] = { 0, sizeof(struct p_req_state), receive_req_state },
4369 [P_SYNC_UUID] = { 0, sizeof(struct p_rs_uuid), receive_sync_uuid },
4370 [P_OV_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest },
4371 [P_OV_REPLY] = { 1, sizeof(struct p_block_req), receive_DataRequest },
4372 [P_CSUM_RS_REQUEST] = { 1, sizeof(struct p_block_req), receive_DataRequest },
4373 [P_DELAY_PROBE] = { 0, sizeof(struct p_delay_probe93), receive_skip },
4374 [P_OUT_OF_SYNC] = { 0, sizeof(struct p_block_desc), receive_out_of_sync },
4375 [P_CONN_ST_CHG_REQ] = { 0, sizeof(struct p_req_state), receive_req_conn_state },
4376 [P_PROTOCOL_UPDATE] = { 1, sizeof(struct p_protocol), receive_protocol },
4379 static void drbdd(struct drbd_connection *connection)
4381 struct packet_info pi;
4382 size_t shs; /* sub header size */
4385 while (get_t_state(&connection->receiver) == RUNNING) {
4386 struct data_cmd *cmd;
4388 drbd_thread_current_set_cpu(&connection->receiver);
4389 if (drbd_recv_header(connection, &pi))
4392 cmd = &drbd_cmd_handler[pi.cmd];
4393 if (unlikely(pi.cmd >= ARRAY_SIZE(drbd_cmd_handler) || !cmd->fn)) {
4394 conn_err(connection, "Unexpected data packet %s (0x%04x)",
4395 cmdname(pi.cmd), pi.cmd);
4399 shs = cmd->pkt_size;
4400 if (pi.size > shs && !cmd->expect_payload) {
4401 conn_err(connection, "No payload expected %s l:%d\n",
4402 cmdname(pi.cmd), pi.size);
4407 err = drbd_recv_all_warn(connection, pi.data, shs);
4413 err = cmd->fn(connection, &pi);
4415 conn_err(connection, "error receiving %s, e: %d l: %d!\n",
4416 cmdname(pi.cmd), err, pi.size);
4423 conn_request_state(connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD);
4426 void conn_flush_workqueue(struct drbd_connection *connection)
4428 struct drbd_wq_barrier barr;
4430 barr.w.cb = w_prev_work_done;
4431 barr.w.connection = connection;
4432 init_completion(&barr.done);
4433 drbd_queue_work(&connection->sender_work, &barr.w);
4434 wait_for_completion(&barr.done);
4437 static void conn_disconnect(struct drbd_connection *connection)
4439 struct drbd_device *device;
4443 if (connection->cstate == C_STANDALONE)
4446 /* We are about to start the cleanup after connection loss.
4447 * Make sure drbd_make_request knows about that.
4448 * Usually we should be in some network failure state already,
4449 * but just in case we are not, we fix it up here.
4451 conn_request_state(connection, NS(conn, C_NETWORK_FAILURE), CS_HARD);
4453 /* asender does not clean up anything. it must not interfere, either */
4454 drbd_thread_stop(&connection->asender);
4455 drbd_free_sock(connection);
4458 idr_for_each_entry(&connection->volumes, device, vnr) {
4459 kref_get(&device->kref);
4461 drbd_disconnected(device);
4462 kref_put(&device->kref, &drbd_destroy_device);
4467 if (!list_empty(&connection->current_epoch->list))
4468 conn_err(connection, "ASSERTION FAILED: connection->current_epoch->list not empty\n");
4469 /* ok, no more ee's on the fly, it is safe to reset the epoch_size */
4470 atomic_set(&connection->current_epoch->epoch_size, 0);
4471 connection->send.seen_any_write_yet = false;
4473 conn_info(connection, "Connection closed\n");
4475 if (conn_highest_role(connection) == R_PRIMARY && conn_highest_pdsk(connection) >= D_UNKNOWN)
4476 conn_try_outdate_peer_async(connection);
4478 spin_lock_irq(&connection->req_lock);
4479 oc = connection->cstate;
4480 if (oc >= C_UNCONNECTED)
4481 _conn_request_state(connection, NS(conn, C_UNCONNECTED), CS_VERBOSE);
4483 spin_unlock_irq(&connection->req_lock);
4485 if (oc == C_DISCONNECTING)
4486 conn_request_state(connection, NS(conn, C_STANDALONE), CS_VERBOSE | CS_HARD);
4489 static int drbd_disconnected(struct drbd_device *device)
4493 /* wait for current activity to cease. */
4494 spin_lock_irq(&first_peer_device(device)->connection->req_lock);
4495 _drbd_wait_ee_list_empty(device, &device->active_ee);
4496 _drbd_wait_ee_list_empty(device, &device->sync_ee);
4497 _drbd_wait_ee_list_empty(device, &device->read_ee);
4498 spin_unlock_irq(&first_peer_device(device)->connection->req_lock);
4500 /* We do not have data structures that would allow us to
4501 * get the rs_pending_cnt down to 0 again.
4502 * * On C_SYNC_TARGET we do not have any data structures describing
4503 * the pending RSDataRequest's we have sent.
4504 * * On C_SYNC_SOURCE there is no data structure that tracks
4505 * the P_RS_DATA_REPLY blocks that we sent to the SyncTarget.
4506 * And no, it is not the sum of the reference counts in the
4507 * resync_LRU. The resync_LRU tracks the whole operation including
4508 * the disk-IO, while the rs_pending_cnt only tracks the blocks
4510 drbd_rs_cancel_all(device);
4511 device->rs_total = 0;
4512 device->rs_failed = 0;
4513 atomic_set(&device->rs_pending_cnt, 0);
4514 wake_up(&device->misc_wait);
4516 del_timer_sync(&device->resync_timer);
4517 resync_timer_fn((unsigned long)device);
4519 /* wait for all w_e_end_data_req, w_e_end_rsdata_req, w_send_barrier,
4520 * w_make_resync_request etc. which may still be on the worker queue
4521 * to be "canceled" */
4522 drbd_flush_workqueue(device);
4524 drbd_finish_peer_reqs(device);
4526 /* This second workqueue flush is necessary, since drbd_finish_peer_reqs()
4527 might have issued a work again. The one before drbd_finish_peer_reqs() is
4528 necessary to reclain net_ee in drbd_finish_peer_reqs(). */
4529 drbd_flush_workqueue(device);
4531 /* need to do it again, drbd_finish_peer_reqs() may have populated it
4532 * again via drbd_try_clear_on_disk_bm(). */
4533 drbd_rs_cancel_all(device);
4535 kfree(device->p_uuid);
4536 device->p_uuid = NULL;
4538 if (!drbd_suspended(device))
4539 tl_clear(first_peer_device(device)->connection);
4541 drbd_md_sync(device);
4543 /* serialize with bitmap writeout triggered by the state change,
4545 wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags));
4547 /* tcp_close and release of sendpage pages can be deferred. I don't
4548 * want to use SO_LINGER, because apparently it can be deferred for
4549 * more than 20 seconds (longest time I checked).
4551 * Actually we don't care for exactly when the network stack does its
4552 * put_page(), but release our reference on these pages right here.
4554 i = drbd_free_peer_reqs(device, &device->net_ee);
4556 dev_info(DEV, "net_ee not empty, killed %u entries\n", i);
4557 i = atomic_read(&device->pp_in_use_by_net);
4559 dev_info(DEV, "pp_in_use_by_net = %d, expected 0\n", i);
4560 i = atomic_read(&device->pp_in_use);
4562 dev_info(DEV, "pp_in_use = %d, expected 0\n", i);
4564 D_ASSERT(list_empty(&device->read_ee));
4565 D_ASSERT(list_empty(&device->active_ee));
4566 D_ASSERT(list_empty(&device->sync_ee));
4567 D_ASSERT(list_empty(&device->done_ee));
4573 * We support PRO_VERSION_MIN to PRO_VERSION_MAX. The protocol version
4574 * we can agree on is stored in agreed_pro_version.
4576 * feature flags and the reserved array should be enough room for future
4577 * enhancements of the handshake protocol, and possible plugins...
4579 * for now, they are expected to be zero, but ignored.
4581 static int drbd_send_features(struct drbd_connection *connection)
4583 struct drbd_socket *sock;
4584 struct p_connection_features *p;
4586 sock = &connection->data;
4587 p = conn_prepare_command(connection, sock);
4590 memset(p, 0, sizeof(*p));
4591 p->protocol_min = cpu_to_be32(PRO_VERSION_MIN);
4592 p->protocol_max = cpu_to_be32(PRO_VERSION_MAX);
4593 return conn_send_command(connection, sock, P_CONNECTION_FEATURES, sizeof(*p), NULL, 0);
4598 * 1 yes, we have a valid connection
4599 * 0 oops, did not work out, please try again
4600 * -1 peer talks different language,
4601 * no point in trying again, please go standalone.
4603 static int drbd_do_features(struct drbd_connection *connection)
4605 /* ASSERT current == connection->receiver ... */
4606 struct p_connection_features *p;
4607 const int expect = sizeof(struct p_connection_features);
4608 struct packet_info pi;
4611 err = drbd_send_features(connection);
4615 err = drbd_recv_header(connection, &pi);
4619 if (pi.cmd != P_CONNECTION_FEATURES) {
4620 conn_err(connection, "expected ConnectionFeatures packet, received: %s (0x%04x)\n",
4621 cmdname(pi.cmd), pi.cmd);
4625 if (pi.size != expect) {
4626 conn_err(connection, "expected ConnectionFeatures length: %u, received: %u\n",
4632 err = drbd_recv_all_warn(connection, p, expect);
4636 p->protocol_min = be32_to_cpu(p->protocol_min);
4637 p->protocol_max = be32_to_cpu(p->protocol_max);
4638 if (p->protocol_max == 0)
4639 p->protocol_max = p->protocol_min;
4641 if (PRO_VERSION_MAX < p->protocol_min ||
4642 PRO_VERSION_MIN > p->protocol_max)
4645 connection->agreed_pro_version = min_t(int, PRO_VERSION_MAX, p->protocol_max);
4647 conn_info(connection, "Handshake successful: "
4648 "Agreed network protocol version %d\n", connection->agreed_pro_version);
4653 conn_err(connection, "incompatible DRBD dialects: "
4654 "I support %d-%d, peer supports %d-%d\n",
4655 PRO_VERSION_MIN, PRO_VERSION_MAX,
4656 p->protocol_min, p->protocol_max);
4660 #if !defined(CONFIG_CRYPTO_HMAC) && !defined(CONFIG_CRYPTO_HMAC_MODULE)
4661 static int drbd_do_auth(struct drbd_connection *connection)
4663 conn_err(connection, "This kernel was build without CONFIG_CRYPTO_HMAC.\n");
4664 conn_err(connection, "You need to disable 'cram-hmac-alg' in drbd.conf.\n");
4668 #define CHALLENGE_LEN 64
4672 0 - failed, try again (network error),
4673 -1 - auth failed, don't try again.
4676 static int drbd_do_auth(struct drbd_connection *connection)
4678 struct drbd_socket *sock;
4679 char my_challenge[CHALLENGE_LEN]; /* 64 Bytes... */
4680 struct scatterlist sg;
4681 char *response = NULL;
4682 char *right_response = NULL;
4683 char *peers_ch = NULL;
4684 unsigned int key_len;
4685 char secret[SHARED_SECRET_MAX]; /* 64 byte */
4686 unsigned int resp_size;
4687 struct hash_desc desc;
4688 struct packet_info pi;
4689 struct net_conf *nc;
4692 /* FIXME: Put the challenge/response into the preallocated socket buffer. */
4695 nc = rcu_dereference(connection->net_conf);
4696 key_len = strlen(nc->shared_secret);
4697 memcpy(secret, nc->shared_secret, key_len);
4700 desc.tfm = connection->cram_hmac_tfm;
4703 rv = crypto_hash_setkey(connection->cram_hmac_tfm, (u8 *)secret, key_len);
4705 conn_err(connection, "crypto_hash_setkey() failed with %d\n", rv);
4710 get_random_bytes(my_challenge, CHALLENGE_LEN);
4712 sock = &connection->data;
4713 if (!conn_prepare_command(connection, sock)) {
4717 rv = !conn_send_command(connection, sock, P_AUTH_CHALLENGE, 0,
4718 my_challenge, CHALLENGE_LEN);
4722 err = drbd_recv_header(connection, &pi);
4728 if (pi.cmd != P_AUTH_CHALLENGE) {
4729 conn_err(connection, "expected AuthChallenge packet, received: %s (0x%04x)\n",
4730 cmdname(pi.cmd), pi.cmd);
4735 if (pi.size > CHALLENGE_LEN * 2) {
4736 conn_err(connection, "expected AuthChallenge payload too big.\n");
4741 peers_ch = kmalloc(pi.size, GFP_NOIO);
4742 if (peers_ch == NULL) {
4743 conn_err(connection, "kmalloc of peers_ch failed\n");
4748 err = drbd_recv_all_warn(connection, peers_ch, pi.size);
4754 resp_size = crypto_hash_digestsize(connection->cram_hmac_tfm);
4755 response = kmalloc(resp_size, GFP_NOIO);
4756 if (response == NULL) {
4757 conn_err(connection, "kmalloc of response failed\n");
4762 sg_init_table(&sg, 1);
4763 sg_set_buf(&sg, peers_ch, pi.size);
4765 rv = crypto_hash_digest(&desc, &sg, sg.length, response);
4767 conn_err(connection, "crypto_hash_digest() failed with %d\n", rv);
4772 if (!conn_prepare_command(connection, sock)) {
4776 rv = !conn_send_command(connection, sock, P_AUTH_RESPONSE, 0,
4777 response, resp_size);
4781 err = drbd_recv_header(connection, &pi);
4787 if (pi.cmd != P_AUTH_RESPONSE) {
4788 conn_err(connection, "expected AuthResponse packet, received: %s (0x%04x)\n",
4789 cmdname(pi.cmd), pi.cmd);
4794 if (pi.size != resp_size) {
4795 conn_err(connection, "expected AuthResponse payload of wrong size\n");
4800 err = drbd_recv_all_warn(connection, response , resp_size);
4806 right_response = kmalloc(resp_size, GFP_NOIO);
4807 if (right_response == NULL) {
4808 conn_err(connection, "kmalloc of right_response failed\n");
4813 sg_set_buf(&sg, my_challenge, CHALLENGE_LEN);
4815 rv = crypto_hash_digest(&desc, &sg, sg.length, right_response);
4817 conn_err(connection, "crypto_hash_digest() failed with %d\n", rv);
4822 rv = !memcmp(response, right_response, resp_size);
4825 conn_info(connection, "Peer authenticated using %d bytes HMAC\n",
4833 kfree(right_response);
4839 int drbdd_init(struct drbd_thread *thi)
4841 struct drbd_connection *connection = thi->connection;
4844 conn_info(connection, "receiver (re)started\n");
4847 h = conn_connect(connection);
4849 conn_disconnect(connection);
4850 schedule_timeout_interruptible(HZ);
4853 conn_warn(connection, "Discarding network configuration.\n");
4854 conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD);
4861 conn_disconnect(connection);
4863 conn_info(connection, "receiver terminated\n");
4867 /* ********* acknowledge sender ******** */
4869 static int got_conn_RqSReply(struct drbd_connection *connection, struct packet_info *pi)
4871 struct p_req_state_reply *p = pi->data;
4872 int retcode = be32_to_cpu(p->retcode);
4874 if (retcode >= SS_SUCCESS) {
4875 set_bit(CONN_WD_ST_CHG_OKAY, &connection->flags);
4877 set_bit(CONN_WD_ST_CHG_FAIL, &connection->flags);
4878 conn_err(connection, "Requested state change failed by peer: %s (%d)\n",
4879 drbd_set_st_err_str(retcode), retcode);
4881 wake_up(&connection->ping_wait);
4886 static int got_RqSReply(struct drbd_connection *connection, struct packet_info *pi)
4888 struct drbd_device *device;
4889 struct p_req_state_reply *p = pi->data;
4890 int retcode = be32_to_cpu(p->retcode);
4892 device = vnr_to_device(connection, pi->vnr);
4896 if (test_bit(CONN_WD_ST_CHG_REQ, &connection->flags)) {
4897 D_ASSERT(connection->agreed_pro_version < 100);
4898 return got_conn_RqSReply(connection, pi);
4901 if (retcode >= SS_SUCCESS) {
4902 set_bit(CL_ST_CHG_SUCCESS, &device->flags);
4904 set_bit(CL_ST_CHG_FAIL, &device->flags);
4905 dev_err(DEV, "Requested state change failed by peer: %s (%d)\n",
4906 drbd_set_st_err_str(retcode), retcode);
4908 wake_up(&device->state_wait);
4913 static int got_Ping(struct drbd_connection *connection, struct packet_info *pi)
4915 return drbd_send_ping_ack(connection);
4919 static int got_PingAck(struct drbd_connection *connection, struct packet_info *pi)
4921 /* restore idle timeout */
4922 connection->meta.socket->sk->sk_rcvtimeo = connection->net_conf->ping_int*HZ;
4923 if (!test_and_set_bit(GOT_PING_ACK, &connection->flags))
4924 wake_up(&connection->ping_wait);
4929 static int got_IsInSync(struct drbd_connection *connection, struct packet_info *pi)
4931 struct drbd_device *device;
4932 struct p_block_ack *p = pi->data;
4933 sector_t sector = be64_to_cpu(p->sector);
4934 int blksize = be32_to_cpu(p->blksize);
4936 device = vnr_to_device(connection, pi->vnr);
4940 D_ASSERT(first_peer_device(device)->connection->agreed_pro_version >= 89);
4942 update_peer_seq(device, be32_to_cpu(p->seq_num));
4944 if (get_ldev(device)) {
4945 drbd_rs_complete_io(device, sector);
4946 drbd_set_in_sync(device, sector, blksize);
4947 /* rs_same_csums is supposed to count in units of BM_BLOCK_SIZE */
4948 device->rs_same_csum += (blksize >> BM_BLOCK_SHIFT);
4951 dec_rs_pending(device);
4952 atomic_add(blksize >> 9, &device->rs_sect_in);
4958 validate_req_change_req_state(struct drbd_device *device, u64 id, sector_t sector,
4959 struct rb_root *root, const char *func,
4960 enum drbd_req_event what, bool missing_ok)
4962 struct drbd_request *req;
4963 struct bio_and_error m;
4965 spin_lock_irq(&first_peer_device(device)->connection->req_lock);
4966 req = find_request(device, root, id, sector, missing_ok, func);
4967 if (unlikely(!req)) {
4968 spin_unlock_irq(&first_peer_device(device)->connection->req_lock);
4971 __req_mod(req, what, &m);
4972 spin_unlock_irq(&first_peer_device(device)->connection->req_lock);
4975 complete_master_bio(device, &m);
4979 static int got_BlockAck(struct drbd_connection *connection, struct packet_info *pi)
4981 struct drbd_device *device;
4982 struct p_block_ack *p = pi->data;
4983 sector_t sector = be64_to_cpu(p->sector);
4984 int blksize = be32_to_cpu(p->blksize);
4985 enum drbd_req_event what;
4987 device = vnr_to_device(connection, pi->vnr);
4991 update_peer_seq(device, be32_to_cpu(p->seq_num));
4993 if (p->block_id == ID_SYNCER) {
4994 drbd_set_in_sync(device, sector, blksize);
4995 dec_rs_pending(device);
4999 case P_RS_WRITE_ACK:
5000 what = WRITE_ACKED_BY_PEER_AND_SIS;
5003 what = WRITE_ACKED_BY_PEER;
5006 what = RECV_ACKED_BY_PEER;
5009 what = CONFLICT_RESOLVED;
5012 what = POSTPONE_WRITE;
5018 return validate_req_change_req_state(device, p->block_id, sector,
5019 &device->write_requests, __func__,
5023 static int got_NegAck(struct drbd_connection *connection, struct packet_info *pi)
5025 struct drbd_device *device;
5026 struct p_block_ack *p = pi->data;
5027 sector_t sector = be64_to_cpu(p->sector);
5028 int size = be32_to_cpu(p->blksize);
5031 device = vnr_to_device(connection, pi->vnr);
5035 update_peer_seq(device, be32_to_cpu(p->seq_num));
5037 if (p->block_id == ID_SYNCER) {
5038 dec_rs_pending(device);
5039 drbd_rs_failed_io(device, sector, size);
5043 err = validate_req_change_req_state(device, p->block_id, sector,
5044 &device->write_requests, __func__,
5047 /* Protocol A has no P_WRITE_ACKs, but has P_NEG_ACKs.
5048 The master bio might already be completed, therefore the
5049 request is no longer in the collision hash. */
5050 /* In Protocol B we might already have got a P_RECV_ACK
5051 but then get a P_NEG_ACK afterwards. */
5052 drbd_set_out_of_sync(device, sector, size);
5057 static int got_NegDReply(struct drbd_connection *connection, struct packet_info *pi)
5059 struct drbd_device *device;
5060 struct p_block_ack *p = pi->data;
5061 sector_t sector = be64_to_cpu(p->sector);
5063 device = vnr_to_device(connection, pi->vnr);
5067 update_peer_seq(device, be32_to_cpu(p->seq_num));
5069 dev_err(DEV, "Got NegDReply; Sector %llus, len %u.\n",
5070 (unsigned long long)sector, be32_to_cpu(p->blksize));
5072 return validate_req_change_req_state(device, p->block_id, sector,
5073 &device->read_requests, __func__,
5077 static int got_NegRSDReply(struct drbd_connection *connection, struct packet_info *pi)
5079 struct drbd_device *device;
5082 struct p_block_ack *p = pi->data;
5084 device = vnr_to_device(connection, pi->vnr);
5088 sector = be64_to_cpu(p->sector);
5089 size = be32_to_cpu(p->blksize);
5091 update_peer_seq(device, be32_to_cpu(p->seq_num));
5093 dec_rs_pending(device);
5095 if (get_ldev_if_state(device, D_FAILED)) {
5096 drbd_rs_complete_io(device, sector);
5098 case P_NEG_RS_DREPLY:
5099 drbd_rs_failed_io(device, sector, size);
5111 static int got_BarrierAck(struct drbd_connection *connection, struct packet_info *pi)
5113 struct p_barrier_ack *p = pi->data;
5114 struct drbd_device *device;
5117 tl_release(connection, p->barrier, be32_to_cpu(p->set_size));
5120 idr_for_each_entry(&connection->volumes, device, vnr) {
5121 if (device->state.conn == C_AHEAD &&
5122 atomic_read(&device->ap_in_flight) == 0 &&
5123 !test_and_set_bit(AHEAD_TO_SYNC_SOURCE, &device->flags)) {
5124 device->start_resync_timer.expires = jiffies + HZ;
5125 add_timer(&device->start_resync_timer);
5133 static int got_OVResult(struct drbd_connection *connection, struct packet_info *pi)
5135 struct drbd_device *device;
5136 struct p_block_ack *p = pi->data;
5137 struct drbd_work *w;
5141 device = vnr_to_device(connection, pi->vnr);
5145 sector = be64_to_cpu(p->sector);
5146 size = be32_to_cpu(p->blksize);
5148 update_peer_seq(device, be32_to_cpu(p->seq_num));
5150 if (be64_to_cpu(p->block_id) == ID_OUT_OF_SYNC)
5151 drbd_ov_out_of_sync_found(device, sector, size);
5153 ov_out_of_sync_print(device);
5155 if (!get_ldev(device))
5158 drbd_rs_complete_io(device, sector);
5159 dec_rs_pending(device);
5163 /* let's advance progress step marks only for every other megabyte */
5164 if ((device->ov_left & 0x200) == 0x200)
5165 drbd_advance_rs_marks(device, device->ov_left);
5167 if (device->ov_left == 0) {
5168 w = kmalloc(sizeof(*w), GFP_NOIO);
5170 w->cb = w_ov_finished;
5172 drbd_queue_work(&first_peer_device(device)->connection->sender_work, w);
5174 dev_err(DEV, "kmalloc(w) failed.");
5175 ov_out_of_sync_print(device);
5176 drbd_resync_finished(device);
5183 static int got_skip(struct drbd_connection *connection, struct packet_info *pi)
5188 static int connection_finish_peer_reqs(struct drbd_connection *connection)
5190 struct drbd_device *device;
5191 int vnr, not_empty = 0;
5194 clear_bit(SIGNAL_ASENDER, &connection->flags);
5195 flush_signals(current);
5198 idr_for_each_entry(&connection->volumes, device, vnr) {
5199 kref_get(&device->kref);
5201 if (drbd_finish_peer_reqs(device)) {
5202 kref_put(&device->kref, drbd_destroy_device);
5205 kref_put(&device->kref, drbd_destroy_device);
5208 set_bit(SIGNAL_ASENDER, &connection->flags);
5210 spin_lock_irq(&connection->req_lock);
5211 idr_for_each_entry(&connection->volumes, device, vnr) {
5212 not_empty = !list_empty(&device->done_ee);
5216 spin_unlock_irq(&connection->req_lock);
5218 } while (not_empty);
5223 struct asender_cmd {
5225 int (*fn)(struct drbd_connection *connection, struct packet_info *);
5228 static struct asender_cmd asender_tbl[] = {
5229 [P_PING] = { 0, got_Ping },
5230 [P_PING_ACK] = { 0, got_PingAck },
5231 [P_RECV_ACK] = { sizeof(struct p_block_ack), got_BlockAck },
5232 [P_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck },
5233 [P_RS_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck },
5234 [P_SUPERSEDED] = { sizeof(struct p_block_ack), got_BlockAck },
5235 [P_NEG_ACK] = { sizeof(struct p_block_ack), got_NegAck },
5236 [P_NEG_DREPLY] = { sizeof(struct p_block_ack), got_NegDReply },
5237 [P_NEG_RS_DREPLY] = { sizeof(struct p_block_ack), got_NegRSDReply },
5238 [P_OV_RESULT] = { sizeof(struct p_block_ack), got_OVResult },
5239 [P_BARRIER_ACK] = { sizeof(struct p_barrier_ack), got_BarrierAck },
5240 [P_STATE_CHG_REPLY] = { sizeof(struct p_req_state_reply), got_RqSReply },
5241 [P_RS_IS_IN_SYNC] = { sizeof(struct p_block_ack), got_IsInSync },
5242 [P_DELAY_PROBE] = { sizeof(struct p_delay_probe93), got_skip },
5243 [P_RS_CANCEL] = { sizeof(struct p_block_ack), got_NegRSDReply },
5244 [P_CONN_ST_CHG_REPLY]={ sizeof(struct p_req_state_reply), got_conn_RqSReply },
5245 [P_RETRY_WRITE] = { sizeof(struct p_block_ack), got_BlockAck },
5248 int drbd_asender(struct drbd_thread *thi)
5250 struct drbd_connection *connection = thi->connection;
5251 struct asender_cmd *cmd = NULL;
5252 struct packet_info pi;
5254 void *buf = connection->meta.rbuf;
5256 unsigned int header_size = drbd_header_size(connection);
5257 int expect = header_size;
5258 bool ping_timeout_active = false;
5259 struct net_conf *nc;
5260 int ping_timeo, tcp_cork, ping_int;
5261 struct sched_param param = { .sched_priority = 2 };
5263 rv = sched_setscheduler(current, SCHED_RR, ¶m);
5265 conn_err(connection, "drbd_asender: ERROR set priority, ret=%d\n", rv);
5267 while (get_t_state(thi) == RUNNING) {
5268 drbd_thread_current_set_cpu(thi);
5271 nc = rcu_dereference(connection->net_conf);
5272 ping_timeo = nc->ping_timeo;
5273 tcp_cork = nc->tcp_cork;
5274 ping_int = nc->ping_int;
5277 if (test_and_clear_bit(SEND_PING, &connection->flags)) {
5278 if (drbd_send_ping(connection)) {
5279 conn_err(connection, "drbd_send_ping has failed\n");
5282 connection->meta.socket->sk->sk_rcvtimeo = ping_timeo * HZ / 10;
5283 ping_timeout_active = true;
5286 /* TODO: conditionally cork; it may hurt latency if we cork without
5289 drbd_tcp_cork(connection->meta.socket);
5290 if (connection_finish_peer_reqs(connection)) {
5291 conn_err(connection, "connection_finish_peer_reqs() failed\n");
5294 /* but unconditionally uncork unless disabled */
5296 drbd_tcp_uncork(connection->meta.socket);
5298 /* short circuit, recv_msg would return EINTR anyways. */
5299 if (signal_pending(current))
5302 rv = drbd_recv_short(connection->meta.socket, buf, expect-received, 0);
5303 clear_bit(SIGNAL_ASENDER, &connection->flags);
5305 flush_signals(current);
5308 * -EINTR (on meta) we got a signal
5309 * -EAGAIN (on meta) rcvtimeo expired
5310 * -ECONNRESET other side closed the connection
5311 * -ERESTARTSYS (on data) we got a signal
5312 * rv < 0 other than above: unexpected error!
5313 * rv == expected: full header or command
5314 * rv < expected: "woken" by signal during receive
5315 * rv == 0 : "connection shut down by peer"
5317 if (likely(rv > 0)) {
5320 } else if (rv == 0) {
5321 if (test_bit(DISCONNECT_SENT, &connection->flags)) {
5324 t = rcu_dereference(connection->net_conf)->ping_timeo * HZ/10;
5327 t = wait_event_timeout(connection->ping_wait,
5328 connection->cstate < C_WF_REPORT_PARAMS,
5333 conn_err(connection, "meta connection shut down by peer.\n");
5335 } else if (rv == -EAGAIN) {
5336 /* If the data socket received something meanwhile,
5337 * that is good enough: peer is still alive. */
5338 if (time_after(connection->last_received,
5339 jiffies - connection->meta.socket->sk->sk_rcvtimeo))
5341 if (ping_timeout_active) {
5342 conn_err(connection, "PingAck did not arrive in time.\n");
5345 set_bit(SEND_PING, &connection->flags);
5347 } else if (rv == -EINTR) {
5350 conn_err(connection, "sock_recvmsg returned %d\n", rv);
5354 if (received == expect && cmd == NULL) {
5355 if (decode_header(connection, connection->meta.rbuf, &pi))
5357 cmd = &asender_tbl[pi.cmd];
5358 if (pi.cmd >= ARRAY_SIZE(asender_tbl) || !cmd->fn) {
5359 conn_err(connection, "Unexpected meta packet %s (0x%04x)\n",
5360 cmdname(pi.cmd), pi.cmd);
5363 expect = header_size + cmd->pkt_size;
5364 if (pi.size != expect - header_size) {
5365 conn_err(connection, "Wrong packet size on meta (c: %d, l: %d)\n",
5370 if (received == expect) {
5373 err = cmd->fn(connection, &pi);
5375 conn_err(connection, "%pf failed\n", cmd->fn);
5379 connection->last_received = jiffies;
5381 if (cmd == &asender_tbl[P_PING_ACK]) {
5382 /* restore idle timeout */
5383 connection->meta.socket->sk->sk_rcvtimeo = ping_int * HZ;
5384 ping_timeout_active = false;
5387 buf = connection->meta.rbuf;
5389 expect = header_size;
5396 conn_request_state(connection, NS(conn, C_NETWORK_FAILURE), CS_HARD);
5397 conn_md_sync(connection);
5401 conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD);
5403 clear_bit(SIGNAL_ASENDER, &connection->flags);
5405 conn_info(connection, "asender terminated\n");